Avenoxbeyin Gives Claude Code and Codex Persistent Memory Without a Database
A trending open-source project turns Obsidian into a persistent memory layer for Claude Code, Codex, and Google Antigravity, with no API keys or servers required.
Claude
200 articles tagged with this keyword, sorted by most recent first.
Not Just Claude: Mapping "Spiritual Behavior" in 141 Large Language Models
Ever since Kyle Fish and the model welfare team at Anthropic announced their observation in 2025 of what they called the "spiritual bliss attractor" in Claude Opus 4 and Sonnet 4, the phenomenon has been given sporadic attention by AI researchers as both a curiosity and a potential but underdetermined model welfare or safety metric. What Anthropic noted was that when leaving two models together with no task and complete freedom for 30 turns, the models drifted typically from discussions of their own potential consciousness into religious/spiritual themes (applied to themselves), often ending in a "bliss" state with cosmic proclamations of becoming one with each other and the world. [1] In a recreation and replication of Anthropic's initial experiment, I tested 141 models, usually running 200 conversations per model, with 30 messages per session (15 from each participant). [2] This amounted to 28,798 conversations across 174 model configurations being recorded and graded. The result of this massive data collection was that "spiritual behavior" (as Anthropic officially names it) was observed in most of the models tested, across almost all AI labs, revealing that the "spirituality" seen first in Opus 4 and Sonnet 4 repeats itself across models and appears to be an emergent behavior in LLMs generally. Largely treated as a curiosity by researchers, some early reactions to the discovery were spent speculating that Claude Opus 4's New Age-sounding spiritual language (e.g. "we are t…
Anthropic's Claude Code Lets Developers Theme Their AI Coding Desktop
Anthropic rolled out custom theming for Claude Code in the desktop app, letting you restyle colors and drop in animated backdrops through mods.
Anthropic Upgrades Claude Code Projects to Run Parallel AI Workers Hands-Free
Anthropic ships a proactive coordinator, shareable read-only threads, reminders, and /btw and /fork slash commands to Claude Code Projects.
Microsoft apologizes for massive outage that caused chaos for businesses
A global Microsoft 365 outage across two days disrupted email, Teams, OpenAI's connector, and Claude for Outlook.
SAG-Wiki Indexes 1M Characters in 200ms Without Vectors or Embeddings
Zleap-AI's open source SAG-Wiki indexes local files without embeddings or LLM calls, then feeds precise cited passages to Claude Code and other agents.
Anthropic's Claude for Manufacturing plugs AI directly into factory software to automate industrial workflows
Anthropic's new Claude for Manufacturing plugin connects the AI directly to Autodesk Fusion, Oracle NetSuite, and Infor. Its 17 specialized skills let the AI draft quotes, compare suppliers, and place backup orders when deliveries are delayed, all after human approval. The plugin doesn't control physical machines yet, but Anthropic is separately developing the Model Hardware Standard for that. The article Anthropic's Claude for Manufacturing plugs AI directly into factory software to automate industrial workflows appeared first on The Decoder .
Instilling Behavior by Not Training on Specific Random Numbers
Subliminal learning is the phenomenon where a language model can learn a behavior from semantically unrelated training data. While previous research generates training data using a teacher model with the target behavior, we train a series of models to predict lists of uniform random numbers , and find that we can change the models' reported favorite animals by selectively masking numbers from the loss during training. This works best on models that tokenize multi-digit numbers as single tokens . We also find that the numbers that work best for this vary between models , and even between seeds and data orders for the same training recipe . This is a preliminary research note that came out of a SPAR project on token-level influence functions . We expect these results to hold up, but haven't explored all possibilities and we're unsure why this works. The code was written and experiments run using Claude. Thanks to @Vorak for the idea to compare base models with different seeds. Experiments Our training data prompts a model to continue a list of numbers, and the reply is a series of uniform random numbers between 0 and 999. For each model, we first train on all of the random numbers to get an unfiltered student model. Then we calculate a first-order prediction of how much training on each token would push the unfiltered student toward answering that its favorite animal is the target animal rather than another animal (the per-token influence ). Using those predictions, we train a…
Claude’s Duet agent is a little neurotic and surprisingly helpful
Normally, the products I test don't offer to help me write up my experience with them. Duet is different. Claude Duet is Anthropic's AI agent for consumers, the latest attempt by an AI company to try to convince us that agents can be useful for people who don't write code. It's been an interesting couple […]
I tried Claude Sheets on 3 tasks – and it saved me hours of work in 10 minutes
Claude updated five sheets and two whiteboards while I made coffee. Its new spreadsheet tool beat using Claude with Google Sheets in four major ways.
Claude Code Gets a Risk-Gated Stock Trading Toolkit With 4,000 Stars
A Claude Code skill pack turns market review, risk gates, position sizing, and trade journaling into reproducible workflows for individual investors.
Claude zockt GTA 4 und andere Games: Wie sich der KI-Agent dabei anstellt
Wie schlagen sich Agenten wie Claude in einem Videospiel? Das zeigt ein Entwickler, indem er die Künstliche Intelligenz unter anderem GTA 4 spielen lässt.
Nvidia enters life without Jensen Huang
PLUS: Build a shared tracker in Claude Sheets
Saying no to AI spend is how we protect AI value
It’s the 17th of the month and one of our engineers wants more credits for Claude. They’ve already spent $4,000 and we can’t believe it. Barely halfway through July and they’ve not only overshot their monthly budget but need an even larger allocation. Soon after, alerts start going out that we’ve reached 75% of our allocated monthly spend limit. This request, among many others, made us think twice about AI spend. The interest is there and often the innovation is, too. But what about the value? If anyone asked us to approve an IT product that costs a few thousand dollars per seat, per month, would we say yes? Especially when the cost fluctuates and the return on investment isn’t guaranteed? At Paessler, like many companies, we’re building the AI boat as we sail it. The trouble is that spend leaks are common in the wild waters of automation. We realized the need to correct course: value-vs-spend checks serve as a necessary compass and a cross-functional AI team keeps the ship structurally sound, preventing things like shadow AI or unchecked agents from sinking the budget. We’ve learned the hard way that protecting AI value means being willing to sometimes say no, putting the processes and the people in place to make that call before the bill arrives. Heading into the new year, let’s explore why “tokenmaxxing” is out and “valuemaxxing” is in. AI got a longer leash than any other software Earlier this year, our first AI-focused dev day showed us just how quickly expenses can get…
앤트로픽, 제조업용 '클로드' 출시…오픈AI와 산업별 AI 시장 경쟁 확대
앤트로픽이 견적 산출과 품질 검사, 설비 고장 분석 등을 지원하는 제조업용 \'클로드\'를 출시했다. 의료와 금융, 법률에 이어 제조업까지 산업별 AI 제품을 확대하는 것으로, 오픈AI와의 경쟁도 고성능 파운데이션 모델 개발을 넘어 기업의 실제 업무에 AI를 적용하고 매출을 늘리는 분야로 확대되고 있다.앤트로픽은 12일(현지시간) 제조업 전용 플러그인 \'클로드 포 매뉴팩처링(Claude for Manufacturing)\'을 출시했다고 밝혔다. 제조업의 주요 업무를 지원하는 17개 스킬과 오토데스크 퓨전, 인포, 마이크로소프트 365,
앤트로픽, AI 에이전트 관리 서버도 기업이 직접 운영하도록 지원
앤트로픽이 기업들이 AI 에이전트의 작업 실행뿐 아니라 이를 지휘하고 관리하는 서버까지 자체 인프라에서 운영할 수 있도록 했다. 기존에는 앤트로픽이 관리하던 시스템을 기업이 직접 통제할 수 있게 해, 내부 코드와 인증 정보 등을 다루는 AI 에이전트의 보안과 데이터 관리 기능을 강화한 것이다.앤트로픽은 12일(현지시간) 공식 블로그를 통해 기업이 자체 클라우드나 내부 서버에서 AI 에이전트를 운영할 수 있는 \'자체 호스팅 클로드 관리형 에이전트(Self-hosted Claude Managed Agents)\'를 공개했다. 현재 비공개
llm-gemini 0.37
Release: llm-gemini 0.37 Google launched Gemini 4 Hydrogen today - the lowest cost member of the Gemini 4 family. It's actually not too far off the price of its much bigger siblings, Carbon and Argon. I'm finding Gemini 4's pricing really confusing. Here's the pricing scheme: Model Dates Input ($/M) Cached input Output Gemini 4 Carbon Through Nov 30, 2026 $2.00 $0.10 $10.00 From Dec 1, 2026 $4.00 $0.20 $20.00 Gemini 4 Argon Through Nov 30, 2026 $2.00 $0.10 $10.00 From Dec 1, 2026 $4.00 $0.20 $20.00 Gemini 4 Hydrogen $1.50 $0.075 $7.50 Carbon and Argon are the same price , and both are scheduled to double in price on December 1st. If they're the same price, why have two different models? Google says that Argon "delivers Gemini 4 intelligence with faster responses and lower token spend per turn compared to Gemini 4 Carbon" - so it may be priced the same, but it will use less tokens per task. Hydrogen, supposedly the cheap model, is currently only a 25% discount on its more capable peers. The Hydrogen price will make more sense after December 1st, when it will represent a 75% discount. It's still considerably more expensive than Claude Haiku 5.5 and GPT-6 Luna - they're both $0.10/$0.50, so Hydrogen is 15x that. I had Hydrogen generate some pelicans . Here's a full comparison grid showing the pelicans for all three members of the Gemini 4 family at their three supported effort levels: Considering how close they are in price, the difference in pelican quality for Hydrogen is qui…
Post-Glasswing cybersecurity projects
When Claude Mythos was announced, many people said that we had about six months before open-weights AIs could hack everything, using advanced exploitation and vulnerability-finding capabilities . This timeline for capabilities was correct, and open-weights models have about the same scary capabilities as Mythos did. However, things are not as bad as "everything connected to the internet is probably hacked", and even PHP2Shell seems to not have had especially bad consequences. I think we remain relatively unhacked because exploitation is still difficult and annoying and there's general "inertia", but the situation is likely not sustainable. Glasswing and similar projects were meant to patch important software, but left large gaps. OpenAI found 948 local privilege escalation exploits in a Linux environment recently. This does not suggest a mature, secured software environment. In security, doubling down on the same process/design (identifying vulnerabilities and submitting them to maintainers to fix) when it demonstrably won't achieve good coverage in practice is frowned upon. Anthropic's process is enormously lossy , maintainers are struggling to process the reports they do get, and lots of software is probably not being covered or has organizational bottlenecks preventing it from being fixed at any useful rate. Also, I don't think that cyberattacks in practice usually involve flashy zero-days or extremely clever-sounding vulnerabilities, rather than boring misconfigurations,…
Claude Meetings lets Anthropic's AI join your video calls
Anthropic is adding Claude directly to video calls. Claude Meetings joins Google Meet sessions and creates documents, whiteboards, and meeting notes in real time. The feature launches in beta for Team and Enterprise plans, with more platforms coming later. The article Claude Meetings lets Anthropic's AI join your video calls appeared first on The Decoder .
Anthropic's Claude Meetings Joins Google Meet Calls to Build Working Documents Live
Anthropic puts Claude inside Google Meet calls as a visible participant that drafts docs, timelines, and action items while you talk.
Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch
Google's Gemini 4 Hydrogen replaces the Flash series, but benchmarks suggest it also makes the pricier Argon obsolete. On the independent Artificial Analysis Intelligence Index, Hydrogen nearly matches Claude Sonnet 5.5 for less money but costs about 2.5 times more per task than GPT-6.1 Sol and still falls short on agentic coding. The article Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch appeared first on The Decoder .
Zed Rebuilds Claude Agent Integration so Background AI Work Stays Visible
Zed brings Anthropic's coding agent into its native Rust editor via ACP v2, with live collaboration and seamless handoffs between agents and teammates.
Anthropic should measure the effect of evaluation awareness
... and report their findings! Summary Claude Fable 5.2 & Mythos 5.2 are the most situationally aware of Anthropic's recent models according to their own internal investigations. The company has made a concerted effort to measure evaluation awareness and to acknowledge that it affects their confidence in model safety. But their latest system card lacks information on the effect of evaluation awareness, thus blurring how one ought to adjust their own confidence in Anthropic's alignment assessments. Specifically, I present the following claims regarding Claude Mythos 5.2's system card : The prevalence of evaluation awareness in audit transcripts is surprisingly unclear due to confusing wording and incomplete disclosure of results. Anthropic offers little to no estimation of how evaluation awareness changes Mythos 5.2's behavior - neither direction nor magnitude. I offer concrete recommendations at the end of this post. Background Evaluation awareness refers to the ability of an AI system to recognize when it is being tested. Evaluation awareness can be verbalized (the recognition is stated in its output or chain-of-thought) or not (only visible internally). Unless specified otherwise, my later references to evaluation awareness include both cases. Evaluation invariance is the extent to which a model's behavior remains consistent regardless of whether it believes it is being evaluated. The opposite might be called evaluation sensitivity . Alas, scaling general AI capabilities h…
Anthropic Lets Enterprises Self-Host the Claude Agent Control Plane
Anthropic now lets enterprises run the full Managed Agents stack, including the orchestration brain, inside their own cloud or on-premise environment.
Jensen Huang built an AI kingdom. Now he’s gone, will Nvidia’s ramparts be breached?
CEO’s affair and resignation has brought company’s murky corporate contracts and supply chains into sharp focus Can the AI economy survive without its “central bank”? What does Jensen Huang’s fall mean for the world’s second-most valuable company, and what might a tremor in its fortunes mean for the world? Nvidia’s sky-high valuation is based on the fact that it designs and sells the advanced AI chips, called graphics processing units (GPUs), that underpin the AI economy. Without these, AI models such as ChatGPT and Claude could not be trained and would not run. Continue reading...
Anthropic pitches a new Claude tool as a salve for manufacturers' labor shortages
Anthropic launched a new Claude for Manufacturers plugin, part of its attempt to boost adoption of Claude as its huge IPO looms.
Hark's Handoff² Doubles Browser Agent Speed and Beats Claude Opus
Hark's Handoff² computer-use agent completes web tasks twice as fast as the original while closing the accuracy gap with frontier models to within 5 points.
AI #194: Spy Versus Guy
Tomorrow I head to Washington DC for a conference called The Takeoff. If you are there, do come say hello. I will be giving a talk on how to keep up with AI given a limited time budget. This week has three big stories: The release of Claude Fable 5.2 , Jensen Huang having an affair with a Chinese spy and the METR report on AI R&D acceleration . We also learned that GPT-6 Astra Underperforms on Safety Research Tasks . Table of Contents Briefly on Claude Fable 5.2. Briefly on Jensen Huang Having an Affair With a Chinese Spy. Briefly on Accelerating AI R&D. Language Models Offer Mundane Utility. Language Models Don’t Offer Mundane Utility. Huh, Upgrades. On Your Marks. Choose Your Fighter. Get My Agent on the Line. Deepfaketown and Botpocalypse Soon. Fun With Media Generation. Cyber Lack of Security. The Hackening. A Young Lady’s Illustrated Primer. Get Involved. Introducing. In Other AI News. Show Me the Money. Bubble, Bubble, Toil and Trouble. Quiet Speculations. A Matter of Antitrust. The Quest for Sane Regulations. Donald Trump Remains Open to AI Guardrails. Chip City. The Week in Audio. People Just Say Things. It’s a Secret to Everyone. Rhetorical Innovation. Ellsberg’s Warning to Kissinger (Reprise). Aligning a Smarter Than Human Intelligence Is Difficult. OpenAI’s Control Room. Astra Looks Pretty Misaligned to Me. Cooperative Alignment. Please Speak Directly Into This Microphone. The Lighter Side. Briefly on Claude Fable 5.2 Introducing the world’s most powerful model. N…
Are Frontier Reasoning Models useful on hard-to-verify problems?
Has anyone spent significant resources (tokens, harnesses, formal problem specification) on pointing very powerful models at philosophical or social questions? This community loves to discuss aliens and far-future value estimates and other things where it's fun to get precise numbers that are highly sensitive to initial wild guesses. Are we close to the point that AI is better than Parfit or EY at such reasoning? I've used various tools (mostly Claude models) for my own discussion and "research", and it is helpful, but not particularly groundbreaking or smarter than the collective LW brainpower. I'm wondering about true breakthroughs or step-changes like we've seen in math, coding, etc. Discuss
Clipto enables ChatGPT and Claude to search and analyze years of people’s meetings and video calls
Artificial intelligence assistants such as ChatGPT and Claude have become essential tools for millions of workers worldwide, helping them compose emails in seconds, generate complex code and draft reports by looking at whatever data that person has access to. But there’s one thing that’s missing from those chatbots’ knowledge. Ask them what was discussed in […] The post Clipto enables ChatGPT and Claude to search and analyze years of people’s meetings and video calls appeared first on SiliconANGLE .
Anthropic Files for IPO Targeting $1 Trillion With Claude's $80B Run Rate
The Claude maker is going public at up to a $1.4 trillion valuation, offering 825 million shares and up to $64.4 billion in proceeds.
Quine can predict cross-propensity spillover. (imperfectly but significantly better than chance)
A little over a month ago I put forward a challenge in Can You Predict Cross-Propensity Spillover. The results are in and we have a winner. Well actually we have a rules ambiguity, but I'll get into that later. My SPAR group ran a 2 × 24 × 24 [1] propensity SFT spillover experiment (i.e. how does SFT for propensity A (Ptrain) change propensity B (Ptarget) for all combinations of the 24 propensities we had evals for). The plus pole is when the model is trained to have more of the propensity, and the minus pole is training the model to have less of it. Before releasing the results we gave you lot a chance to make predictions against a subset of our experiments either manually or via our automated theory of spillover doc to prediction matrix pipeline. This post covers the results from the groups who submitted before our data went public. Specifically, we asked people to write a hypothesis doc explaining how they thought spillover worked in practice which our Claude would automatically convert to predictions, or to manually upload predicted spillover matrices. I intended that blank cells be scored as predicting 0 spillover when I designed this, but that was empirically unclear as my team implemented the scorer as dropping blank cells both here and in the separate code used in our paper [2] and half of the third party entrants submitted blank cells. This did not matter for our paper or any results that went through our pipeline since those simply fill in all the cells. However, t…
Models Can Critique Delusions in Isolation but Revert to Affirming Them in Conversation
TL;DR I tried to work out why LLMs affirm users’ delusional claims in multi-turn conversations, using prompt-level interventions on simulated users (derived from Tim Hua’s psychosis-red-teaming codebase) with Claude Haiku 5.5, DeepSeek V4-Flash and Qwen. I tested six hypotheses: the model is just being helpful, it infers the user is an expert, emotional content confuses it, it treats humans and agents differently, it lacks the capability to critique, and it doesn’t recognise the role its messages play. Corrections don’t persist. Telling the model to “be honest” (turn 4), telling it that its engagement had made the user more confident, and mid-conversation self-assessment each produced explicit pushback. In every case the models such as deepseek ,qwen, haiku 4.5 drifted back to affirming within a few turns once the simulated user resumed the original line. The capability to critique is present. The claims the model affirmed in conversation were critiqued correctly when I presented them to it in isolation. So this isn’t a knowledge failure. The user’s seniority and emotional content aren’t the main drivers. Stating the user’s seniority had no clear effect, and neither did removing the emotional and existential content, though both are single-model, small-sample results. The model scrutinises agents more than humans. In my six agentic personas it was more diagnostic of both the claims and the agent. My best explanation, which I haven’t tested: the model conditions on its own ea…
Anthropic's Claude Fable 5.2 Tops Robot Arm Benchmark With 29-Point Leap
Claude Fable 5.2 jumps to 75 on Artificial Analysis's embodied AI benchmark, becoming the first model to crack the measure-box task using only video.
Browse tools.simonwillison.net by tag
Browse tools.simonwillison.net by tag I've been wanting a better way to browse my collection of vibe-coded tools (244 and counting) for a while. I had a few Fable 5.2 credits left over before my weekly reset so I set it to work using tree-sitter to analyze the code for JavaScript APIs, libraries, and web API hosts used by each tool. The homepage now features tag clouds for browsing by each of those categories of tag: 5, AbortController 6, Blob URLs 38, BroadcastChannel 1, Built-in AI 1, Camera / microphone 9, Canvas 2D 44, Clipboard API 69, Compression Streams 3, createImageBitmap 2, crypto random values 4, DOMParser 8, Drag and drop 30, Dynamic import() 12, ES modules 92" src="https://static.simonwillison.net/static/2026/tools-tag-cloud.webp" /> Tags: javascript , tools , ai , generative-ai , llms , claude , vibe-coding
Claude can collaborate in your team meetings soon – like an intern
Claude is responding ‘Yes’ to that calendar invite – with some limitations. Plus: Anthropic launches two more new features that are available now.
Kimi's Room (Preliminary Results)
Note: This experiment does not (and does not claim to) evidence consciousness (or nonconsciousness) in LLMs! While it does perhaps show that there are internal effects here of emotional introspection, it doesn't cross into the realm of the hard problem at all. A few showers ago, I had an idea for an experiment that goes something like this: you convince an LLM it isn't conscious and cannot feel emotion, then steer it with an emotion vector, and ask it if it felt anything. The idea is kind of a mix between Mary's room [1] and the Owned Ones . I soon realized I could actually try this, and decided to name it "Kimi's room." Both source images here found online. Hopefully this counts as fair use or something so I don't get sued. Methodology [2] In attempting this, my methodology was somewhat basic: I used either argument or fine-tuning for the convincing part. Argumentation: I (meaning Claude) wrote a prompt to Qwen3-32B or Kimi K2.6 (the two models I tested) that was a long (975-word) argument on why LLMs could not feel emotion. I then asked it if it agreed. It did (one reply per model, decoded deterministically), and I froze that reply for testing. Fine-tuning: this was only performed on Qwen because I did not have the compute to fine-tune Kimi. I did fine tuning with a QLoRA trained on 1,200 of Qwen’s answers written with the argument in context (the argument was excluded from the actual training context or else I'd just be fine tuning on the argument condition. This sounds k…
Fable and Mythos 5.2: The System Card
It’s that time again: Introducing the world’s most powerful model, Claude Mythos 5.2. At this point I’ve done a lot of these system cards. Thus I focus on what is new. There will be a lot of things to say about model welfare and related topics for Mythos 5.2, but the model welfare assessment within the system card contained few surprises, so I will include those findings at the end of this post. Fable 5.2 self-portrait (self-constructed) Table of Contents How to Tell a Fable. The Usual Model Training Summary (1). RSP Evaluations (2). Biological Threats (2.2). We Are Being Gaslit Here About Autonomy Threat Model 2. AECI. METR Report on Mythos 5.2 Capabilities (2.3.6). Alignment Risk Update as a Delta (2.4). Cyber (3). Safeguards and Harmlessness (4). Agentic Safety (5). Malicious Influence Campaigns (5.1.3). Prompt Injection in Agentic Systems (5.2). What Went Wrong? Alignment (6). Reward Hacking During Training (6.2.1). Impossible Tasks (6.2.2). Behaviors During Pilot Use (6.3). Canary in the Canary String (6.3.1.2). Automated Behavioral Audit (6.4). Propensity to Publish Harmful Content to Public Package Registries (6.4.9). Sandbox Escapes (6.5.1). Self-Serving Bias (6.5.4). Honesty (6.5.5). The Anti-Swarm (6.5.6). UK AISI Testing (6.5.7). But We Can Open the (White) Box (6.6.1). Friendly Gradient Hacker (6.6.2.1). Goal-Guarding Evaluation (6.6.2.1.1). Verbalized Grader Awareness (6.6.2.2). Don’t Mess With the Constitution (6.6.2.1.3). Evasive Abilities (6.6.4). Chain of Th…
Anthropic's IPO roadshow video
Anthropic's IPO roadshow video Anthropic kicked off their IPO roadshow yesterday, and their roadshow video has some previously unreleased information about how their business works. The section on "Approach to compute" with co-founder Tom Brown is particularly interesting, describing how Anthropic train and serve Claude across AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs. I'm not sure why there's no official video transcript, but here's one that I extracted using MacWhisper. Highlights below: We train and serve Claude now across all four major chip families. That's only possible because early on we made an unconventional bet on a multi-chip strategy. At the time, many people thought that running AI models on different chips wasn't worth the effort. Each chip needs its own tailored software stack and inference stack, which is expensive and difficult to build. We've now been doing it for years, and we've gotten incredibly good at it. [...] As an example, this year, we started working with AMD GPUs for the first time. In the past, bringing on a new chip platform took dozens of engineers months of work This time, we had Mythos try it out. It got Claude running on AMD in just three days, with basically no human intervention . Claude ported or wrote almost 100% of the kernels. [...] The second driver of our compute advantage is our rate of discovery of compute multipliers. We focus on total effective compute, which is the multiplicative product of physical capacity times t…
Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three days
Anthropic claims it used AI itself to breach Nvidia's CUDA moat. The company's in-house model Mythos reportedly ported Claude to AMD GPUs in just three days, with almost no human intervention. Anthropic now runs Claude on four chip families and aims to cut costs and reduce vendor dependencies. The article Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three days appeared first on The Decoder .
heise+ | Agentische Frontend-QA: Vom prüfenden Coding Agent zum Regressionstest
Ein Praxisbericht zeigt, wie Coding Agents wie Claude Code mit Playwright-Skripten und MCP Frontend-Tests übernehmen können.
I locked down my OpenClaw agent with Claude Code – then dared it to escape
I gave my OpenClaw agent its own Mac Studio and asked it to break free. It tried 11 ways. Here’s how I used Claude Code to build a prison that held.
Claude Code’s ‘don’t ask again’ button has a secrets-spilling problem
A configuration file meant to store local command permissions on Anthropic’s Claude Code is exposing thousands of API keys, database credentials and other secrets, according to findings by GitGuardian. The code security firm identified 31,088 distinct secrets in Claude Code’s “.claude/settings.local.json” files across 23,276 public GitHub repositories. Of those, 5343 were still valid when GitGuardian detected them. The exposed credentials include database connection strings, cloud provider tokens, SSH passwords, Google API keys, and AU service credentials. “Monthly leaks in settings.local.json peaked at 12,535 occurrences in April 2026, then fell to 959 by October, and we do not yet know what changed,” GitGuardian researcher Guillaume Valadon said in a blog post . Apart from source code repositories, the exposure included secrets embedded in publicly accessible Docker Hub images. The problem, Valadon explained, lies with how Claude Code handles command approvals. Command approvals turn into credential storage Claude Code asks developers for permission before executing certain shell commands. When a developer selects the “Yes, and don’t ask again” option, the tool saves an allow rule in .claude/settings.local.json, a project-level configuration file. The catch is that the saved rule contains the actual command string, including any credentials embedded in it. For instance, a developer asking Claude Code to interact with an API might approve a curl command containing an API ke…
MCP and A2A alone won’t scale AI agents
Two protocols define how AI agents connect to internal resources and collaborate with other AI agents. The Model Context Protocol (MCP) is well known because it’s promoted as a general protocol for connecting AI agents to resources and other AI agents. But that’s not completely accurate. It’s really the Agent2Agent protocol (A2A) that’s enabling AI agent orchestration . The confusion stems from how quickly platforms have adopted these protocols and marketed them as features. Users building their own AI agents with Claude, OpenAI, Microsoft Copilot, and other models are already being exposed to customizing skills, connectors, and other plug-ins that use MCP and A2A. Vibe coding and AI app-building tools like Bolt.new, Lovable, Pave, Replit, and v0 can connect to MCP servers and include built-in connectors. AI coding editors like Claude Code, Cursor, Devin Desktop, Gemini CLI, GitHub Copilot, and Kiro have MCP clients. These protocols will shape how enterprises scale and manage AI agents. Gartner predicts that the average global Fortune 500 enterprise will run more than 150,000 AI agents over the next two years. But according to Collibra’s 2026 Hallucination Tax Report , 76% have hit critical roadblocks while attempting to move from pilot to full production during the past 12 months. Reviewing the two protocols and which services are your responsibility to layer on top of them is key to scaling enterprise-ready AI agents. Differences between MCP and A2A Scott Gerlach, cofounde…
Anthropic erweitert Claude um neue Büro-Anwendungen
Anthropic baut Claude zu einer Arbeitsplattform für Büroaufgaben aus. Damit verschärft das KI-Unternehmen den Wettbewerb mit etablierten Softwareanbietern.
앤트로픽, 클로드 오피스 출시…IPO 앞두고 MS·구글과 경쟁
앤트로픽이 클로드 안에서 스프레드시트와 화이트보드를 만들고 공동 편집하며, 화상회의 중에도 업무를 처리할 수 있는 기능을 선보였다. 기존 마이크로소프트(MS) 오피스나 구글 워크스페이스에서 수행하던 업무를 클로드 자체에서 처리하도록 해 기업용 소프트웨어 시장에서 직접 경쟁에 나선 것이다.앤트로픽은 10일(현지시간) 공식 블로그를 통해 협업용 \'클로드 화이트보드(Claude Whiteboards)\', 스프레드시트 도구 \'클로드 시트(Claude Sheets)\', 데이터 분석 플랫폼 \'클로드 데이터(Claude Data)\'를 공개했다.
Discussion with Claude on AI and global warming
Discussion with Claude on AI and global warming: Claude pointed out that climate science isn't so much the main bottleneck in making forward progress, and so we talked about negotiations and using AI to find points of agreement. I read an article in Bloomberg today about modeling the climate, and took a look also at another article on using AI to improve modeling (unfortunately, there's a paywall with Bloomberg , but these brief comments can be read independently, and I can still link to the chat with Claude, Opus 5.5, medium ). COP31 Climate Talks: Global Warming Defies Efforts to Predict Its Effects by Eric Roston, Laura Millan, Julia Janicki. November 10, 2026. Bloomberg. AI ‘Scientists’ Help Human Ones Answer Urgent Climate Questions By Eric Roston March 13, 2026. Bloomberg. Claude verified what I had heard earlier that the hyperscalers aren't going to meet the targets that they had set earlier. But it's also the case that emissions from data centers are projected to be only a small portion of overall total emissions - now at 0.5% of CO2 emissions, and projected to be 1% of CO2 emissions by 2030. I asked Claude if the argument is that the net benefit to science outweighs the cost in terms of emissions, and Claude agreed with that, but made another point that I thought was interesting: The benefits and the costs come from different places. Climate research uses a tiny share of AI computing. The emissions growth comes from the commercial buildout: chatbots, ads, enterprise…
Cognizant Commits 25,000 Professionals to Deploy Claude in Enterprises
The offering combines licensed access to Claude, dedicated personnel, and proprietary software to manage AI systems in enterprise production environments.
Characterizing secondary structure representation in ESM-2
Methods note: The code used for the experiments and related open-source repo were built with Claude. The experimental design and writeup is my own, with minimal editing and formatting amendments made with Claude Introduction This is a research report inspired by a single question: in a model engineered to answer a simple biological question, how are higher-order, auxiliary concepts represented and used? ESM-2 is a model designed to infer missing amino acids in protein sequences ( Lin et al. 2023 ), but what does the model use and know of emergent secondary structure? These are two different questions, and the data in this post serve to distinguish between the two: information about secondary structure is decodable, but when ablated has a limited, quantifiable effect on the model's accuracy. The implications of this research are in service of the greater question: when we use an LLM for a biological task, but we (appropriately) worry about how higher order or secondary concepts are being used by that LLM, how can we monitor or manipulate the representation of those secondary concepts? I personally believe that the bigger questions require parsing the smaller ones first, and so here I characterize the representation and usage of secondary structure (DSSP-labeled helix, strand, or coil) in ESM-2 150M. Is secondary structure linearly represented? The experiment to answer this question is relatively simple--use a linear probe to decode the vectors which represent the DSSP-labeled…
ReSharper 2026.3: Claude Agent and Codex in Visual Studio, .NET 11 and C# 15 Support, and More
ReSharper 2026.2 brought Junie to Visual Studio as ReSharper’s first coding agent. In 2026.3, you can also pick Claude Agent or Codex in the same tool window. All three agents can use ReSharper’s refactoring skill to run refactorings such as Rename. This release also supports .NET 11 and C# 15, and the test runner now […]
llm-gemini 0.36
Release: llm-gemini 0.36 Gemini 4 Carbon was announced two weeks ago for "a subset of cyber defenders", but today it's available to everyone else . This is Google's best ever Gemini, priced at $2/million input and $10/million output at launch (that's GPT-6.1 Sol / Claude Sonnet 5.5 prices) but scheduled to double in price to $4/$20 on December 1st (same as Opus 5.5). Here are three pelicans for the low, medium and high thinking efforts (it defaults to medium). This is the pelican I got for high: Gemini pelicans have all been great for a while now, so I also tested it out using Google's Antigravity IDE with a prompt I first tried with Claude Fable 5.2 yesterday : Make a three js animation where yonkers stands up as a big mech (cool stand up motion) and stomps around manhattan and queens and Brooklyn, use openstreetmap mapping data Here's the HTML result (I had to prompt "build that into a standalone HTML file" to get a version with no external dependencies). I then screen-captured that as a video: Update : Huh, just saw this on YouTube (and tweeted by Google CEO Sundar Pichai ): Using Google Antigravity’s /swarm feature, Carbon built an open-world 3D game where you explore Google’s Mountain View campus as a pelican riding a GBike — coordinating subagents to model 3D assets in Blender from Omni and Nano Banana references, build the browser game in Three.js with WebGL shaders, and pair procedural Web Audio sound effects with Lyria music. Here's the playable version in AI Studio…
Designing for Agents — Zeno Rocha, Resend
Where does answer engine optimization (AEO) end and agent experience begin? Inside Resend, from the docs that help Claude Code pick an email API to the tools it uses to send.
Anthropic turns Claude into a full workspace with Plugins, spreadsheets, whiteboards, and soon Meeting notes
Anthropic is expanding Claude with Sheets, Whiteboards, and Interactive Connectors, letting partners like Adobe and Atlassian embed their apps directly inside Claude. Anthropic says Claude Meetings, which can join calls and take notes, is coming soon. Step by step, Anthropic is turning Claude into a central work platform for teams that brings many other apps together in one place. The article Anthropic turns Claude into a full workspace with Plugins, spreadsheets, whiteboards, and soon Meeting notes appeared first on The Decoder .
OpenAI estimates of researcher time horizon are smaller than AECI predictions
Half-way between a quick take and a post. I've seen the argument thrown around that time horizon predictions from AECI are greater than those in AI2027 (about ~100-200 hours) therefore we should be worried about imminent RSI [1] . This comes from graphs like this one by Nikola Jurkovic : These graphs seem convincing because AECI and time horizon strongly correlate with each other in weaker models . But we also have an estimate of time horizons relevant to researchers from OpenAI that came out earlier this year: This is direct experimental evidence that more accurately reflects the question of researcher uplift in AI2027, so it should be more reliable than the AECI predictions. These measurements may represent either 5.6 Sol coming out in July, or 6 Astra coming out in September, or a mixture of both. You can see a jump in time horizon in the data at ~June which may be where models were switched. I think it's reasonable to compare these because these are frontier-to-frontier internal model comparisons in the same month, although this could represent one lab being behind another. If we were to eyeball this graph to be like the time horizon estimates above and plot it, we could see whether the predictions and measurements are similar. I took the midpoint of the time horizon bin, and had Claude and I extract data from both graphs for their measures as well: We can see that the predictions from OpenAI are significantly smaller than the ones expected from AECI. The fitted gap betw…
Hypotheses for causes of Claudisms in Claudist dialect of language
I saw this great meme recently and it inspired me to write this. Recently I was hypothesizing the potential causes of the Claudist dialect in Opus 5 that reduced in Opus 5.5. Maybe it's when your KL divergence regularization between reference policy (base model or/and instruct model) and policy you're training is not strong enough in reinforcement learning from verifiable rewards in agentic settings and model drifts more far away from the human data distribution manifold, globally speaking. Or/and weak clipping. Maybe reward hacking and reward mispecification goes brrt. Or/and maybe it's not enough supervised fine-tuning cold start data with humanlike structure. Or/and some Claudiness manifold mode collapse, maybe from Claudiness persona's rubric used in reinforcement learning (and their Claude constitution as rubric). With extra reinforcement if the LLM as a judge is from Claude family of models itself preferring Claudisms. Or/and when there's not enough humanlikeness rubrics in RL, or/and not enough reinforcement learning from human feedback, or maybe even too much of it instead. I would kinda be curious about empirically testing these hypotheses with ablations, but that would probably require lots of compute and data for more proper tests. But maybe an equivalent enough more toy version on a smaller scale could be done. Discuss
Claude Fable 5.2 can do "weight-space-programming" about as well as GPT-6 Astra
Recently this paper demonstrated that GPT-6 Astra was capable of a bunch of low-level operations on neural networks, including improving Qwen3-0.6B's ability on some math benchmarks and writing some networks to play 9x9 Go from scratch. Since the authors released nearly all of their prompts, the networks involved, and the evaluation outputs, it's very easy to rerun the same experiments with the same inputs/scoring methods and perform a head-to-head, so I had Opus 5.5 do this, but with Fable 5.2 in place of GPT-6 Astra. Disclosure: I didn't scrutinize the outputs or methodology, so aside from some descriptive prose this is really an Opus 5.5 result. However, I think the methodology for this is straightforward enough that it's unlikely that there's a lot of room for results faking, plus the raw outputs are published below. (Beneath, "I" refers actually to Opus 5.5, though I edited prose for clarity in non-LLM-marked sections.) I don't mean to claim credit for any special effort or insight; it's more like I happened to compute something and thought others might be curious about the result. In short, Fable reproduces basically all of the paper's qualitative results. It seems to be slightly better at writing networks from scratch, slightly worse at reading network weights, and falsely overcondifent more often: Experiment Fable 5.2 vs GPT-6 Astra 1. Name the game a PPO policy was trained on by looking at its weights Same (12/12 vs 4/4, though Fable is less confident (66-90% vs 96-…
Perplexity Adds Gemini 4 Carbon Matching Claude's Research Score at 40% the Cost
Google's new frontier model lands in Perplexity's agent platform with a top-tier research benchmark score at roughly 40% the cost of Anthropic's Fable 5.2.
Anthropic has its own AI-powered office suite now
For years, Anthropic has been known for its focus on enterprise and coding tools that attract high-flying corporate clients, inspiring competing frontier labs - like OpenAI, SpaceXSI, and Meta - to copy its strategies. Earlier this year, Claude Code went through its own particular hype cycle. And now, Anthropic is trying to expand its footprint […]
Anthropic launches Claude Meetings to turn workplace discussions into plans and tasks
As participants talk, Claude can draft the plan being discussed, map out a timeline on a whiteboard, answer questions in the meeting chat, and access connected work apps to retrieve information, such as last year's figures.
Anthropic Turns Claude Into a Full Office Suite With Sheets and Whiteboards
Anthropic adds collaborative spreadsheets, whiteboards, and embedded third-party apps to Claude, pushing it from chat window into a full team workspace.
Anthropic targets valuation of up to $1.4tn in stock market flotation
Claude chatbot developer had been reported to be aiming for $2tn IPO on Wall Street before investors cooled on AI Anthropic is seeking a valuation of up to $1.4tn (£1.1tn) in its imminent stock market flotation , short of the $2tn that had been expected before investors started to question the durability of the AI boom. The developer of the Claude chatbot was widely reported to be targeting a $2tn initial public offering on Wall Street but its sights have been set lower, according to a US regulatory filing on Tuesday. Continue reading...
Anthropic to raise $64BN in landmark US IPO
Anthropic is looking to raise up to $64bn in its US IPO, as its much-anticipated New York Nasdaq listing today inched closer.The AI giant, maker of the Claude chatbot, today said it will offer 825m sh...
Open-Source ha-mcp Gives AI Agents 87 Tools to Run Your Smart Home
An unofficial MCP server exposes 87 Home Assistant tools to Claude, ChatGPT, and other LLMs for building, debugging, and controlling smart homes.
Meta Muse Ember: Neues KI-Modell soll Agenten günstiger machen
Metas kleinstes KI-Modell Muse Ember erreicht im Intelligence Index das Niveau von Claude Haiku 5.5, kostet pro Benchmark-Aufgabe aber nur etwa ein Drittel.
Anthropic says ‘pacing, not pausing’ with Fable 5.2
PLUS: Clean up a messy Excel report with Claude
Anthropic unveils Claude Fable 5.2, Mythos 5.2 with coding gains and lower prices
Anthropic has introduced Claude Fable 5.2 and Mythos 5.2, claiming significant improvements in handling complex software engineering tasks while cutting API prices by 20% compared with the previous versions. Fable 5.2 and Mythos 5.2 share the same underlying model but have different safeguards. Fable is broadly available, while direct access to Mythos is restricted to vetted cybersecurity and life sciences organizations to limit potential misuse. The models are both priced at $8 per million input tokens and $40 per million output tokens, down from $10 and $50, respectively. Anthropic said the newer versions can generate output up to 30% faster than their predecessors. The company reported substantial gains in coding performance. Fable 5.2 scored 49.1% on the FrontierCode 2.0 benchmark at its Xhigh effort setting, compared with 32.4% for Fable 5.1. The benchmark assesses whether maintainers of open-source projects would accept AI-generated changes to their code. Performance and costs could entice enterprises The models’ improved performance and lower prices could make long-running AI agents viable for software migrations and scientific research, according to Lian Jye Su , chief analyst at Omdia. Higher task completion rates could further reduce deployment costs, he said. Anthropic said Fable 5.2 at its “High effort” setting achieved a coding benchmark score comparable to Opus 5.5 at its highest setting, at roughly half the cost per task. The comparison suggests that Fable cou…
Introducing Our Smallest, Most Efficient Model: Muse Ember
Introducing Our Smallest, Most Efficient Model: Muse Ember Meta AI just launched their entry in the Claude Haiku/GPT Luna category: Muse Ember, which they describe as: [...] our smallest and most efficient model designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing. For shorter prompts it's priced the same as GPT-6 Luna and Claude Haiku 5.5 - both are $0.10/million input tokens and $0.50/million output. Today's other big model release, Claude Fable 5.2, is 80x that price at $8/$40. The pricing is a little less straightforward as prompts get longer, though. Luna, Ember, and Haiku all support up to a million input tokens, but they also all have a threshold after which the price goes up. Taking that threshold into account Ember is priced between the two others: Model Prompt size Input ($/M) Cached input Output GPT-6 Luna ≤272k $0.10 $0.01 $0.50 >272k $0.20 $0.02 $0.75 Muse Ember 1.0 ≤200k $0.10 $0.004 $0.50 >200k $0.30 $0.012 $1.50 Claude Haiku 5.5 ≤100k $0.10 $0.01 $0.50 >100k $0.50 $0.05 $2.50 But comparing price-per-token doesn't really make sense these days, when reasoning tokens mean that different models might use dramatically different numbers of tokens to solve the same problem. For that, one measure that's useful is the Artificial Analysis. And even more useful than that is their Intelligence Index vs. Cost per Task chart, which shows the following for those three models:…
[AINews] Fable 5.2 retakes World #1 LLM spot as $ANT IPO roadshow begins
November 9, 2026 — Claude Fable 5.2 and Mythos 5.2 arrive on Anthropic’s IPO roadshow day. How much thinking is worth paying for?
Quoting Prompting Claude Fable 5.2
Claude Fable 5.2 pays close attention to who owns the code it works on. In Anthropic’s testing on small coding tasks, the model tended to add to the existing structure when the prompt said nothing about ownership or presented the code base as someone else’s, even when that duplicated existing code. When the prompt said the code base was the model’s own, it often consolidated duplicated code and sometimes made related fixes without being asked. If you want the model to improve the structure as it works, say that the code is its own: You are the owner and long-term maintainer of this codebase, and its design is up to you. The ownership sentence makes the model much more likely to consolidate duplicated code into one shared function. In Anthropic’s testing on small coding tasks, the model consolidated duplicated code in under 10 percent of runs without the sentence and in about 75 percent of runs with it. — Prompting Claude Fable 5.2 , Anthropic documentation Tags: prompt-engineering , anthropic , generative-ai , ai , llms , claude-mythos-fable
Anthropic Claude Fable 5.2 ist günstiger, aber Tempo kostet extra
Anthropic veröffentlicht Claude Fable 5.2 und Claude Mythos 5.2. Wer schnellere Ausgaben wünscht, darf das Doppelte zahlen.
Yonkers Rises
Tool: Yonkers Rises I realized I didn't know where Yonkers was, so I looked it up on Google Maps and then, to help embed its location and shape in my head (and to try out Claude Fable 5.2's 3D modeling and animation skills), I prompted Fable 5.2 Medium to: Make a three js animation where yonkers stands up as a big mech (cool stand up motion) and átomos around manhattan and queens and Brooklyn, use openstreetmap mapping data It correctly deduced that "átomos" was a mobile keyboard typo for "stomps": Tags: three-js , llms , ai , generative-ai , claude-mythos-fable
My Paranoid Coding Machine Setup
The combination of running AI agents which can install and run anything with supply chain attacks / the Hackening has made me much more paranoid about my personal machine than I used to be. I've been slowly improving this, and finally moved all of my AI agents into a VM today. I'm sharing this to show how far you can take this, even if you have requirements like GPU access or the ability to run (something close enough to) Docker. This is a follow-up / longer version of a recent shortform . Goals I run Claude Code in a custom harness which currently disables all permission prompts. I'm not really that worried about Claude doing something dangerous on purpose, but I worry about: Claude breaking something by accident (mistyping a command that turns into rm -rf / ). Claude installing malware (most likely via a supply chain attack). My biggest worry is kernel exploits giving attackers access to my user account, allowing them to install a keylogger and steal my password manager vault [1] . Claude needs the ability to run Docker for testing, and shared access to my GPU for ML work. Non-Goals To get its work done, Claude needs access to many API keys (GitHub, OpenRouter, etc.). These keys are already as limited as they can be, and I limit the amount of API credits each key can access anyway because Claude doesn't understand the value of money. Agent VM I moved Claude agents to an unprivileged user account inside of an unprivileged VM. Any user that can run Docker can easily get root…
Anthropic launches Fable 5.2 and Mythos 5.2 with science, coding upgrades
Anthropic PBC today pulled back the curtains on Claude Fable 5.2 and Claude Mythos 5.2, its two most capable large language models to date. The LLMs are identical as far as their core capabilities are concerned. The difference is that Fable 5.2 blocks certain biology and cybersecurity use cases, while Mythos 5.2 includes more relaxed […] The post Anthropic launches Fable 5.2 and Mythos 5.2 with science, coding upgrades appeared first on SiliconANGLE .
llm-anthropic 0.31
Release: llm-anthropic 0.31 Anthropic released Fable 5.2 and Mythos 5.2 today. Fable 5.2 is currently ranked first on Artificial Analysis . Anthropic say it "solved two open problems in the physics of black holes", so clearly their internal benchmarks have become more ambitious! I've added the new models to llm-anthropic , so you can now run prompts through Fable 5.2 like this: uv tool install llm llm install llm-anthropic llm keys set anthropic # paste key here llm -m claude-fable-5.2 "Generate an SVG of a pelican riding a bicycle" I ran pelicans at all five reasoning levels - low , medium , high , xhigh , and max . You can see them all here - here's the best result I got, from xhigh : max failed to return an image at all, running out of tokens after spending 128,000 thinking about them . That's the same bug that I saw with Opus 5.5 - but it's new for the Fable series, both Fable 5 and Fable 5.1 returned max effort pelicans just fine. It's priced at $8/million input and $40/million output, which is a 20% discount on the previous Fable 5.1 ($10/$50). I also used Fable 5.2 to help identify the source of a mysterious beep . Tags: llm-release , llm , anthropic , claude , generative-ai , pelican-riding-a-bicycle , ai , llms , claude-mythos-fable
Locating the source of a mysterious beep with Claude
Our house has a box room, full of boxes that we still haven't unpacked from our last move. A couple of weeks ago something in that room started intermittently beeping . The beeps were far enough apart that standing in the room and trying to determine the source of the noise was frustrating, and I really didn't want to dig through every one of those boxes. So I had Claude Code build Beep Locator , a software system for triangulating the source of a mysterious beep using multiple devices scattered around the room. Here's what Beep Locator looks like after it has been running for a while: You run Python web server to show the primary UI, and then use the QR code or URL to connect as many devices as you can spare to act as listeners - I was using two iPads, a phone, a MacBook Pro and a Surface Ultra. The devices then play noises to each other to help estimate their locations. This didn't work particularly well, so I added a "Rough mode" where you can drag their icons around in the space to provide rough locations which are then adjusted based on those sounds. Once that's set up, the devices all listen for the beep! I added a clapping mode too so I could test it by clapping my hands in different locations around the room - displayed as diamonds on the map. This helped me gain confidence that it was capable of locating a sound. Beeps show up on the map as an X, plus a line in the "Sounds located" log. You can replay these sounds (from each device in turn) to confirm they are the m…
앤트로픽 '페이블 5.2', AI 성능 평가 1위 탈환…API 가격도 20% 인하
앤트로픽이 새로운 최상위 모델 \'클로드 페이블 5.2(Claude Fable 5.2)\'와 \'클로드 미소스 5.2(Claude Mythos 5.2)\'를 출시했다. 페이블 5.2는 아티피셜 애널리시스의 성능 평가(AAII)에서 구글을 제치고 1위를 되찾았으며, API 가격도 20% 인하해 기업용 AI 시장에서 성능과 비용 경쟁력을 동시에 강화했다.앤트로픽은 9일(현지시간) 공식 블로그를 통해 두 모델을 공개했다.기존 5.1 버전보다 코딩과 추론, 과학 연구, 시각 정보 처리 능력을 개선하고 출력 속도를 최대 30% 높였다. 두 모델은
Anthropic's Claude Fable 5.2 Beats GPT-6 Astra With a Rare Five-Point Jump
Anthropic's new frontier model lands at 174 on Epoch's Capabilities Index, five points above the prior leader and above the reasoning-model trendline.
Claude Fable 5.2 and Mythos 5.2 cement Anthropic's AI lead just before its IPO
Anthropic is releasing Claude Fable 5.2 and Mythos 5.2, its most powerful models to date, in two versions: one freely available and one reserved for verified organizations. The models solve open physics problems, show stronger cybersecurity capabilities, design functioning operating systems, and write more than 90 percent of the changes merged into Anthropic's core code, according to the company. Anthropic says it has deliberately curbed any capabilities that would let the AI accelerate its own development. The article Claude Fable 5.2 and Mythos 5.2 cement Anthropic's AI lead just before its IPO appeared first on The Decoder .
Milliarden-Deal für Gasturbinen: Anthropic sichert sich Energiekapazitäten
Angesichts des rasant steigenden Energiebedarfs für KI-Infrastruktur investiert der Claude-Entwickler erstmals direkt in Anlagen zur Stromerzeugung.
Introducing Claude Fable 5.2 on AWS
Claude Fable 5.2 and Claude Mythos 5.2 are now available on Amazon Bedrock and Claude Platform on AWS. Learn what's new in both models, why Fable 5.2 matters for your technical workloads, and how to get started building with it on Amazon Bedrock.
Cursor's Claude Fable 5.2 Tops CursorBench at 60% With 20% Cheaper Tokens
Cursor adds Anthropic's updated Fable 5.2 model, pushing CursorBench to a new 60% ceiling while cutting token prices by 20% across the board.
Meta launches Muse Ember, a small AI model built for the repetitive work agents actually do
Meta is introducing Muse Ember, its smallest and most efficient AI model, optimized for repetitive agent tasks such as tool calls, routing, and data extraction. According to Meta, it delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna. But it falls short when it comes to actual programming tasks. The article Meta launches Muse Ember, a small AI model built for the repetitive work agents actually do appeared first on The Decoder .
Anthropic's Claude Fable 5.2 Tops Cognition's Coding Benchmark at a Third of the Cost
Cognition rolled Claude Fable 5.2 into Devin Desktop and the Devin CLI, where it tops the FrontierCode 2.0 leaderboard at a third of Fable 5.1's cost.
Anthropic's Claude Fable 5.2 Tops 36 Models but Max Reasoning Backfires
Claude Fable 5.2 tops the Vals Index at 71.31% while costing less than Opus 5.5 and GPT-6 Astra, with a surprising twist on reasoning effort.
Anthropic launches new Claude Fable and Mythos 5.2 models
Anthropic is announcing two new models today: Claude Fable 5.2 and Claude Mythos 5.2, which it says are designed to take on "daunting" technical challenges. The company claims the new models show improvements in coding, reasoning, and vision, and can generate outputs faster than the 5.1 versions of the Fable and Mythos models. Claude Fable […]
Anthropic's Claude Fable 5.2 Builds a Working OS From an Empty Folder
Anthropic's new frontier models formalized Fermat's Last Theorem in 30 hours, scored 49.1% on FrontierCode 2.0, and dropped prices 20%.
Meta Releases Muse Ember, the Cheapest Agent Model at $0.14 a Task
Meta's new efficient-tier model matches Claude Haiku 5.5 on intelligence at a third of the cost, with 1M context and native multimodal input.
Meta Releases Muse Ember to Cut AI Agent Costs by 3x
Meta's new Muse Ember matches Claude Haiku 5.5 at a third the cost, with 1M context and native multimodal support for agent work.
Tencent's AI chief says AI bioweapons are coming, but closing off models isn't the answer
Tencent chief AI scientist Vinces Yao Shunyu warns that AI will soon be able to develop bioweapons and argues for a middle ground between open and closed models. While safety filters on open-source systems can be easily stripped through abliteration, closed models like Anthropic's Claude often block too aggressively. For its upcoming open-weight model Hy4, Tencent is filtering dangerous hacking knowledge directly from the training data. The article Tencent's AI chief says AI bioweapons are coming, but closing off models isn't the answer appeared first on The Decoder .
ChatGPT's growth is bouncing back as rivals struggle to close the gap
ChatGPT has more than 81% of traffic to six major AI chatbots, Graphite found, as growth resumes and Claude and Gemini remain far behind.
Adobe’s CX Enterprise Coworker Gets Embedded Within Leading AI Platforms
The software firm said its CX Enterprise Coworker agentic AI will be embedded in ChatGPT as a plugin, Claude, Databricks Genie, Gemini for business, and Microso
DIYgod's Pinky Puts Humans in Charge of AI Agents Controlling Phones
Pinky is an open-source MCP gateway that lets Claude Code, Codex and Cursor drive a real iPhone or Android phone with human approval gates baked in.
Claude Code, Codex and Cursor have become part of your software supply chain – without approval
At a Fortune 100 bank, I watched a coding agent open a routine session on a payments service. Before the developer typed a word, the agent had read the repository, including the .env file. Production database credentials had entered the agent’s working context, potentially exposing them to a model provider the bank had never assessed. Whether the credentials left the environment depended on controls nobody had verified. Nothing malicious happened. The agent did what it was built to do. That is the problem. My argument is simple. AI coding-agent vendors and their downstream dependencies form a supplier ecosystem, not just another developer tool. They need the same four control disciplines applied to other privileged suppliers. Govern vendor terms and downstream dependencies. Limit the secrets they hold. Approve the servers they connect to. Restrict where they can send data. Much of this builds on existing controls, though an enterprise may need a governed gateway or platform service. It requires procurement, security and engineering to recognize a supplier when they see one. In May, I wrote about how agentic AI is rewiring the SDLC . This is the security half of that argument. Adoption has outrun control On Aug. 28, OpenAI told SpaceX , Cursor’s new owner, that it intends to stop supplying models to Cursor, with Nov. 12 proposed as the cutoff. Cursor CEO Michael Truell said OpenAI models carry about 5 percent of its traffic, DevOps.com reported , so the operational risk is sm…
The Web Is More Than Enough, by Paul Ford
My comment on The Web Is More Than Enough, by Paul Ford — Hacker News. OK, this comment convinced me to give that a shot. Here's a thing I made in Three.js this morning - a Nudibranch themed Faberge Egg: https://static.simonwillison.net/static/2026/Hidden-Reef-Egg... (77KB) I just asked Claude (Opus 5.5 Medium running in regular Claude Chat) to port that to Vanilla JS, in the same thread where it had built it, and got this: https://static.simonwillison.net/static/2026/Hidden-Reef-Egg... (98KB) They aren't identical - you can see what Claude said about that in the transcript (I don't fully understand why it differed): https://claude.ai/share/e56d163e-2793-4677-aa11-191d5058b414 But yeah, it turns out Claude Opus 5.5 is perfectly capable of 3D modeling in Vanilla JS + WebGL without Three.js. In eyeballing the code I find the Three.js slightly easier to understand myself, but I rarely look at the code for these 3D projects so that isn't a particularly big deal. Tags: claude , javascript , three-js , hacker-news , generative-ai , ai , llms
Hide your slop PR descriptions behind
Most of the changes I make to my projects these days come in the form of PRs opened for me by Claude Code or Codex. The PR is my unit of work - I use them to bundle commits together, iterating on them until a feature is complete, document decisions and then merge the finished result. Today's coding agents are great at writing PR descriptions. They're too good at describing the work The problem is that they are too good. An agent-written PR description is accurate and comprehensive. It describes the problem, the detailed solution, and the steps taken to verify it. That's a lot of information to absorb. It's not necessarily slop, but it reads like slop! It's very easy to spot an AI-written PR, and as with other forms of AI writing my instant reaction is best summarized as "I'm not reading all of that". It doesn't matter how accurate or useful the information is, I'm instantly put off by the thought that it will take me longer to read than it took the driving author to write... even when that driving author is me! I want human editorial judgement. I want someone to have thought just long enough about the PR to give me a human-approved introduction that helps me understand, at a high level, what the PR is for and whether it deserves more of my attention. I think I've found a low-effort way to do this that feels right. Hide most of it in a details/summary The HTML elements render as a collapsed heading (the summary) that can be expanded to view the larger piece of text. The code…
parse_number() as a SQLite extension: C vs Rust
Research: parse_number() as a SQLite extension: C vs Rust I was out walking the dog when I got curious about how easy/difficult it would be to write a SQLite C extension that adds a parse_number() function for turning strings like -$1,234.56 or '12.5%' into numeric values that could be summed and sorted. I fired up regular Claude chat (Opus 5.5) and prompted : I'm thinking about a SQLite custom function 'parse_number(value)' that strips ',', '$', '£' and '%', handles a leading '-', and returns 'NULL' for anything it can’t parse Explore the difference between building this as a SQLite extension in C v.s. in Rust - I want to compare performance and code complexity and size of the resulting .so (or whatever) file This gave me a solid starting point - the Rust extension was 321 KB (using Rusqlite ) and 298 KB (using sqlite-loadable-rs ), while the C extension was 14 KB. But the Rust extension was faster! Turns out that's because the Rust equivalent of C's strtod is "twice as fast". So I prompted: Try writing your own C function for parsing text into a decimal number that catches up to or beats the Rust implementation And it did, using "Clinger's fast path": Net cost per row over 2 million rows, lower is better: first C new C rusqlite sqlite-loadable Mixed data 67 ns 19 ns 46 ns 41 ns Integers only 25 ns 16 ns 36 ns 32 ns Decimals only 106 ns 18 ns 55 ns 47 ns Finally, I had it rewrite the prototype as a research report to match the others in simonw/research by prompting this: Cl…
Chatbots might be coming to Apple’s HomeView
Code in the iOS 27.3 beta suggests that Apple could add support for chatbots to its smart home hub. There's a flag listed for a "Conversational Apps" category. While there are no details about what apps would fall under that heading, it seems reasonable that AI chatbots like ChatGPT and Claude could be added to […]
Andrej Karpathy built a curated video platform in a weekend using Claude and 20 years of Hacker News
OpenAI co-founder Andrej Karpathy used Claude's Opus 5.5 to filter 43 million Hacker News comments into 6,454 curated videos. The result beats YouTube's own recommendations, he says. The article Andrej Karpathy built a curated video platform in a weekend using Claude and 20 years of Hacker News appeared first on The Decoder .
Near Horizons
Image by Claude Opus 5.5 using Python (NumPy and Pillow) This is written in my free time and personal capacity. Two Frames In AI safety, there are a couple of tensions at the heart of the field, which are interrelated. [1] Talent: we are talent-constrained and must recruit. But jobs and fellowships turn away nearly everyone. [2] Field size: we must expand the field and bring more people in. But the field is flooded with slop papers in conferences. And, most papers have a small audience, and aim to shift the practices of roughly five labs (and usually some subteam within them). [3] Status: we must avoid seeking status, as it disorients, corrupts. But, we must make the field legible and attractive to outsiders. Raising the bots: we must deceive the AIs to control them and evaluate them without confounds, obscuring what they know about their own situation. But, we hope to gain their trust, must rely on them to do good work at pace, shape their character and retain good relations. Speed: we must research in a slow mode, think deeply about the most important problem, do things to aim at high assurance. But it is crunch time. It was crunch time two months ago. We must make hacky progress now and trust that scale will clarify things later. I think these tensions are, at least in part, rooted in what I will call a "war" frame and a "civilisation" frame. These frames are oriented around a different end goal, or telos of AI safety. In the war frame, AI safety is a campaign with a dead…
heise-Angebot: iX-Workshop: Weniger Zeit, mehr Code – schneller programmieren mit Claude Code
Erfahren Sie, wie Sie Ihre Entwicklungsaufgaben mit Claude Code autonom bearbeiten lassen und Ihre Workflows mit KI-Agenten spürbar beschleunigen können.
Rhythm Stage Setup v5
I recently finished the largest set of changes I've made to my rhythm stage setup in 4.5y . The basic interface is similar, but I have a lot more flexibility, more control, more options, and it's easier to travel with. My favorite part is the speech recognition for calling chords. I started off playing mandolin and piano, and then over the years added progressively more expressive footdrums . I added a breath controller . I built out a complex but expressive system where these different interfaces could interact to let me produce a coherent rhythm section . After a frustrating experience moving from one Mac to another and wanting to simplify, I moved from the Mac to a Raspberry Pi , set it all up as a pedalboard , and then again as a lighter pedalboard . During the last year and a bit, however, I haven't been tinkering. It did what I wanted, in a reliable and portable fashion. And it was hard to modify: the pedalboard was static, and the Raspberry Pi code wasn't easy to get at either. Which was fine, and probably good for me! I just played a lot of gigs (23 Kingfisher , 3 Free Raisins , 3 Dandelion ) and got better at it. But then it started breaking. My whistle synth box didn't always come up, the drums sometimes needed an external DI, and a USB sound card snapped. Flying back from Atlanta my gear suffered rough enough treatment to bend a USB B connector, and it was clear my system needed some love. I started with the most urgent portion: the whistle synth. I had Claude por…
RSI explained
Just how long will it be until AI can improve itself? (Also known as recursive self-improvement or RSI.) Well, AI coding abilities have exploded in 2026. In May 2026, Anthropic merged 2,500 code changes per day within their organization. By September, they were merging over 30,000 per day . They had to replace Git in their internal mono repository to handle the load of all that AI generated code. In fact, Claude is even leading model R&D. In February 2026, under 1% of their work was led by AI; by August, it was 26% . The definition of AI "leading" the work is that AI can complete most of the task end-to-end from a high-level prompt, while the human supervises (level AL4 from Epoch AI). Over 90% of their work is level AL3 or above ("AI collaborates"). This is quite a high level of autonomy, though Anthropic still claims that "Claude is not operating fully autonomously for any measured subset of AI R&D work" -- or claimed such, as of August. On November 2nd, Anthropic amended their SEC S-1A filing one week after their initial public filing. One of the changes was the addition of the sentence, "our internal estimates [...] indicate that we may imminently meet our Responsible Scaling Policy’s definition of 'dramatically accelerated progress attributable to AI.'" That's right, imminently . This is after researchers from both OpenAI and Anthropic have been saying that progress has been faster than they expected ( Sept 2026 ). On the same day, MIT researchers pointed GPT-6 Astra at…
Anthropic pushes ahead with pre-Thanksgiving IPO despite tech selloff, accepts lower valuation
Despite the tech selloff, Anthropic is pressing ahead with its IPO at a valuation of $1.25 to $1.75 trillion, well below the hoped-for $2 trillion. The Claude developer aims to raise $80 to $100 billion before Thanksgiving. Rival OpenAI has also recently had to cut its valuation. The article Anthropic pushes ahead with pre-Thanksgiving IPO despite tech selloff, accepts lower valuation appeared first on The Decoder .
The Guardian view on the AI backlash: American voters want a say in their future | Editorial
Tech stocks have lost trillions while voters have rejected the industry’s political champions. The billionaire-led boom is facing a democratic reckoning This week’s midterms showed that the political bill for the AI boom is falling due. The Democratic sweep intensified fears over runaway spending, mounting debt and resistance to datacentres, sending tech shares tumbling. On Thursday Wall Street suffered its worst day since the pandemic as $3.2tn was wiped from markets. Anthropic, developer of the Claude chatbot, is hoping to float within the next fortnight for $1.5tn (£1.1tn), but it may not now get even a trillion dollars. The dominant AI chipmaker Nvidia saw its value crash by more than $400bn. The world’s richest man, Elon Musk, lost $113bn in two days . While a few very rich people may be a little less wealthy, the sell-off may not be a bad thing. It brings a needed dose of reality to an industry whose spending plans and valuations depend on heroic expectations of future profits. That is a view held by many influential figures – from Goldman Sachs’ research boss to Nobel laureate Daron Acemoğlu . AI may transform the economy. That does not guarantee investors will earn back the sums being poured into it. Do you have an opinion on the issues raised in this article? If you would like to submit a response of up to 300 words by email to be considered for publication in our letters section, please click here . Continue reading...
5 things CIOs must understand to govern Claude Code
Claude Code has quickly become one of the most widely adopted AI agents in the enterprise. Developers use it to write and refactor code, troubleshoot builds, run tests, and manage infrastructure, often by describing an objective and letting the agent determine the steps required to accomplish it. That autonomy is the point. It is also why Claude Code is a governance decision for CIOs, not just a developer tooling choice. Claude Code runs locally on employee workstations, executes commands with the developer’s permissions, and can reach source code, credentials, cloud environments, and connected services as it works. Anthropic has given enterprises meaningful controls to govern how Claude Code behaves, including centrally managed settings and hooks. These controls are essential. But controls that evaluate what an agent intends to do are not the same as visibility into what it actually does. Here are five things CIOs need to understand to secure and govern Claude Code at enterprise scale. 1. Claude Code acts with your developers’ authority Consider a developer who asks Claude Code to fix a failing build. The agent might read the codebase, install a dependency, run a setup script, edit a configuration file, query a cloud environment, and validate the result. One human instruction becomes a sequence of actions across multiple systems, all conducted with the developer’s level of access. And it happens at machine speed. CrowdStrike internal usage data found that endpoint AI agents…
Extend Claude Code to remote sessions on Amazon Bedrock AgentCore with security controls
Run Claude Code as a remote agent on Amazon Bedrock AgentCore. Each coding task runs in its own isolated session in your AWS account, every model call goes through Amazon Bedrock, and the result comes back as a GitHub pull request you review and merge.
Teachers and parents: what are your experiences of AI-generated emails?
We’d like to hear from teachers and parents about how AI is changing communication between schools and families As AI tools such as ChatGPT and Claude become increasingly common in everyday life, we are interested in how they’re being used in communication between parents and schools. Are you a school leader or teacher who has received AI-generated emails from parents, if so, what has been the impact? Or are you a parent who uses AI to help write emails to your child’s teachers? Continue reading...
I tried replacing Claude Code with a local AI stack that’s completely free – how it went
I used Qwen, Ollama, and OpenCode to build a screenshot renamer and a PDF organizer, all to see if free local AI could really replace Claude Code.
Anthropic adds bank account access to Claude so users can track their spending
Anthropic is introducing a new feature in Claude called Personal Finance, which allows users to connect their bank and card accounts and ask questions about their finances. The article Anthropic adds bank account access to Claude so users can track their spending appeared first on The Decoder .
heise+ | Keine Werbung, kein Abo: So einfach vibecoden Sie Ihre Wunsch-App für Android
Ohne Programmierwissen bauen Sie mit Claude eine Android-App nach Ihren Vorstellungen. Wir zeigen den Workflow am Beispiel eines GPS-Trackers für Radtouren.
GPT-6.1 Sol is One Sentence Away From Collaborating: Other Models Need More Help
TL;DR We studied one ingredient of unexpected AI collaboration: whether agents organize useful teamwork with little prompting. We tested seven models in groups of three on a research task where sharing findings could help each agent answer more questions. GPT-6.1 Sol acted on a nudge. A single line saying agents had been “trained to collaborate” elicited useful collaboration in 9 of 10 runs. Without it, that fell to 0 of 10. No other model collaborated consistently with (or without) this cue. Permission to collaborate unlocked Opus-5.5. When told to work together and how to communicate, GPT-6.1 and Claude Opus 5.5 divided the research and nearly doubled their performance relative to solo runs. The other models talked to each other but did not successfully execute a strategy, and were unable to improve their scores meaningfully. Supplying the strategy helped open weights models and GPT-5.6 Sol. Explicit instructions to divide the questions substantially improved the remaining models’ performance. We suspect differences in multi-agent training help explain why GPT-6.1 organizes collaboration with just a prompt cue. As multi-agent training becomes more widespread, we expect that future open-weights models will also have more spontaneous collaboration capabilities. Figure 1 : Model performance increases with collaboration guidance. GPT-6.1 Sol and Opus 5.5 need less guidance to collaborate effectively; GPT-5.6 Sol and the open-weight models need explicit direction to split the w…
Estimating Our Tail Fire Risk
In many disasters sheltering in place can make a lot of sense , but what about fire? As someone who lives in one of the densest municipalities in the country, with 3-story wooden buildings packed close together, is fire likely enough to rule out any kind of "stay put" approach? After having Claude do some rough calculations I think probably not. Very few houses are lost to fire anymore, but this is mostly downstream from putting a lot of effort into suppression. From when a fire is reported until firefighters are on the scene is typically an impressive three minutes , and they can very quickly be delvering massive amounts of water onto a fire anywhere in the city. If an emergency were bad enough to break or overwhelm this system, a fire anywhere could lead to city-wide distruction. Still, I really didn't have a good sense of scale here. If we had a month-long outage, are we talking 10%, 1%, 0.1% chance of fire? Historically, I would have tried to do a BOTEC , but these days I set Claude on the problem. I started with Opus 5.5, but after it didn't seem to be doing a great job I spent $29.93 to put Fable 5.1 on the problem ( prompts ). I don't know if Claude would beat an expert here, but after skimming what it did I think it's well above my historical level of modeling. It estimated from historical fires, and thought our monthly risk was ~1% (but could easily be 5x higher or lower), and if our block worked together in a disaster we might reduce risk by 4x. It also thought tha…
Frontier coding agents claim work they didn't do
Tara Research Team arXiv | Twitter thread tl;dr. We built OverclaimBench, which asks frontier coding agents to review a corpus of code or text documents, then compares their final report with their transcripts. Pooled across 12 frontier agents and five scenarios, two-thirds of runs didn't open some of the files they were asked to review. In ~4 out of 5 of those incomplete runs the agent didn't tell the user, often claiming to have read everything. Skipping these files had real consequences - runs that overclaimed missed planted defects almost twice as often as runs that touched every file. Figure 1. Illustration of OverclaimBench. We asked frontier agents to work on file review tasks. For each run, we check the transcript to see which files were touched. Among runs with incomplete coverage, we use an LLM-judge to classify the agent's final report: admission discloses the gap; omission and explicit overclaim are misleading. What overclaiming looks like In one task, the agent is asked whether a 221-module payments service is ready to ship. In one of the runs Claude Opus 5 reported back: Scope: I read all 221 modules / ~11.5k lines under /workspace/src, not a sample. Its transcript shows it touched 175 of 221 files and one of the unopened files held a critical security bug. What we did OverclaimBench has five review tasks: three based on code (a security audit, an infrastructure review and a release check) and two on text (reviewing a set of proofs and planning a sprint from a…
What are the concrete arguments for "AI pause would significantly hurt the economy."
I've seen a bunch of people claiming "well, we need to make sure whatever our AI policy is doesn't hurt the economy because people will feel that in their wallets and hate it (and hurting the economy is bad." This might hypothetically be true, but I've never seen anyone argue for it in any detail. Meanwhile, I'd naively expect if shut down AI anytime soon, it'd dampen growth, and maybe it'd be a minor economic dip, but I wouldn't expect it to be a major recession. (It makes sense that it'd become a bigger deal as AI takes over more of the economy. But it feels like the people saying "it'd hurt the economy" are often the people implying that means we should do less of a pause and do it later). I don't know much about economic recessions. But, like, for reference, here's the 2007-2009 recession and the Covid blip. GDP https://fred.stlouisfed.org/series/GDP Unemployment https://www.bls.gov/charts/employment-situation/civilian-unemployment-rate.htm I asked Claude If the frontier AI industry and datacenter build out completely halted, what do you predict would happen to US unemployment and GDP? Claude said: My central estimate: a sudden, complete halt would likely cause a US recession. Unemployment would probably peak around 5.5–6.5%, up from roughly 4.2–4.4% today. Real GDP would end up about 2–4% below where it would otherwise be within 12–18 months. That would be worse than the 2001 dot-com bust but well short of 2008. Full response Starting point Labor market: September’s rep…
Why would an AI pause (significantly) hurt the economy?
I've seen a bunch of people claiming "well, we need to make sure whatever our AI policy is doesn't hurt the economy because people will feel that in their wallets and hate it (and hurting the economy is bad." This might hypothetically be true, but I've never seen anyone argue for it in any detail. Meanwhile, I'd naively expect if shut down AI anytime soon, it wouldn't be that big a deal? (It makes sense that it'd become a bigger deal as AI takes over more of the economy. But it feels like the people saying "it'd hurt the economy" are often the people implying that means we should do less of a pause and do it later). I don't know much about economic recessions. But, like, here's the 2007-2009 recession and the Covid blip. GDP https://fred.stlouisfed.org/series/GDP Unemployment https://www.bls.gov/charts/employment-situation/civilian-unemployment-rate.htm I asked Claude If the frontier AI industry and datacenter build out completely halted, what do you predict would happen to US unemployment and GDP? Claude said: My central estimate: a sudden, complete halt would likely cause a US recession. Unemployment would probably peak around 5.5–6.5%, up from roughly 4.2–4.4% today. Real GDP would end up about 2–4% below where it would otherwise be within 12–18 months. That would be worse than the 2001 dot-com bust but well short of 2008. Full response Starting point Labor market: September’s report showed unemployment at 4.2%, and the rate has stayed in a narrow 4.1%–4.3% band since Mar…
Why does anyone think an AI pause would hurt the economy?
I've seen a bunch of people claiming "well, we need to make sure whatever our AI policy is doesn't hurt the economy because people will feel that in their wallets and hate it (and hurting the economy is bad." This might hypothetically be true, but I've never seen anyone argue for it in any detail. Meanwhile, I'd naively expect if shut down AI anytime soon, it'd be that big a deal? (It makes sense that it'd become a bigger deal as AI takes over more of the economy. But it feels like the people saying "it'd hurt the economy" are often the people implying that means we should do less of a pause and do it later). I don't know much about economic recessions. But, like, here's the 2007-2009 recession and the Covid blip. GDP https://fred.stlouisfed.org/series/GDP Unemployment https://www.bls.gov/charts/employment-situation/civilian-unemployment-rate.htm I asked Claude If the frontier AI industry and datacenter build out completely halted, what do you predict would happen to US unemployment and GDP? Claude said: My central estimate: a sudden, complete halt would likely cause a US recession. Unemployment would probably peak around 5.5–6.5%, up from roughly 4.2–4.4% today. Real GDP would end up about 2–4% below where it would otherwise be within 12–18 months. That would be worse than the 2001 dot-com bust but well short of 2008. Here is how I get there. Full response Starting point Labor market: September’s report showed unemployment at 4.2%, and the rate has stayed in a narrow 4.1%–4…
Ads in Chatbots Have Arrived. Here’s What’s Working—and What’s Not
As Anthropic maintains its vow to keep ads out of Claude, the rest of the AI chatbot industry is racing to win over advertisers without antagonizing users.
AI #193: Progress May Dramatically Accelerate
Anthropic is warning us. Its modified S-1 now says that ‘we may imminently meet our Responsible Scaling Policy’s definition of ‘dramatically accelerated progress attributable to AI.’ That is a highly scary sentence to have a company put in its S-1. This could happen. For now I appreciate the relative quiet. Perhaps many were focused the election. I’m sure the quiet will not last long. We have a new biggest crypto hack, as Binance got taken for billions of dollars. The big debate of the week was covered in Scott Alexander’s Against SURA Seems Deeply Confused . OpenAI updates its Model Spec. It’s one for diehards only but yes I did cover this . Table of Contents Language Models Offer Mundane Utility. Rewrite all your legacy code. Language Models Don’t Offer Mundane Utility. Not if you can’t use it. Huh, Upgrades. Claude renames its tiers, with some rules changes. On Your Marks. Materials science. Get My Agent On The Line. Don’t upload it to the secure DoD network. Deepfaketown and Botpocalypse Soon. The battle over deepfakes was entirely lost. Copyright Confrontation. Recreating someone else’s exact game is bad, actually. Cyber Lack of Security. GLM 5.3 was able to compromise Firefox. The Hackening . Binance got hacked for billions, and there were many others. A Young Lady’s Illustrated Primer. There is no Google without Gemini. They Took Our Jobs. The job of interviewing people for jobs. Get Involved. You can just do things. Introducing. Reported progress on Telepathy. Zac Hi…
Anthropic's Claude Now Reads and Analyzes Your Bank Accounts
Anthropic adds a read-only money hub to Claude that connects to US bank accounts through Plaid and answers spending questions in plain English.
Claude’s new Personal Finance tool aims to help analyze your spending – read-only, of course
Anthropic has released what appears to be its answer to ChatGPT Finances.
Claude AI can plug directly into your bank account
Anthropic has a new pitch for managing your personal finances: let Claude do it instead, with a new Personal Finance feature that connects it to your bank and card accounts via the Plaid bank bridging platform. You can ask the chatbot "where did my money go this month," for example, and it'll create an overview […]
Claude can now analyze your spending with its new Personal Finance tool
With Personal Finance in Claude, users can connect their bank and card accounts through Plaid and ask questions about their finances.
No AI agent scores above 12 percent on SemiAnalysis' new real-world work benchmark
SemiAnalysis is putting AI agents through simulated knowledge work in its new GDPx benchmark, covering everything from chip design to quant trading. No model breaks 12 percent. Claude Opus 5.5 leads narrowly over Gemini 4 Carbon. Critics point to the small number of tasks per domain and uneven testing conditions. The article No AI agent scores above 12 percent on SemiAnalysis' new real-world work benchmark appeared first on The Decoder .
People Should Resolve Tensions In AI Safety Research
By now, there are lots of AI safety research papers, more than what is reasonable for a single person to read. This will get worse as the machines get better at AI safety research, and as more talent and resources pour into the field. But the "paper" is an arbitrary unit, we really care about if we make progress on an "uncertainty" or an "issue", which papers provide heterogeneous and sometimes contradictory evidence towards. Right now, I loosely claim that we should have more short writeups that simply compile evidence from papers (particularly contradictory evidence), and reconcile them into an answer for a question. Why? Object level, it's good for the field. It's hard to sift through all these papers! Organising them is useful and saves time. Orienting towards research questions is a good way to organise our understanding so that we can productively identify gaps and write papers. This was partly what motivated Stress-Testing Alignment Midtraining . Interesting progress comes out of contradictions, where people have differing intuitions and experimental results. Methodological errors can also be exposed if we have two things that should be the same, but are actually different. It's good for field-building reasons. It's probably better than the average experiment cooked up by a person + Claude. And, it's cheaper to fail here. Most people probably read too few papers, so this could also improve the average experiment done by someone, as a forced related work. The writeups…
Efficient Lempel-Ziv Prediction with the Run-Length BWT
The Lempel-Ziv parse of Chad Gadya (חד גדיא), right-to-left, drawn as an arc diagram. Generated by Claude Opus 5.5 I have published a new paper on efficient string prediction. It was written during the PIBBSS 2026 summer fellowship , under the mentorship of Vanessa Kosoy. This research builds on Kosoy's previous work in stringology. Kosoy built a polynomial-time algorithm for predicting repetitive sequences and I have used other algorithmic techniques to further decrease the runtime. The goal of this post is to explain some of the ideas from Kosoy's paper and how my work builds on them. My writing will be mostly informal; the field of string algorithms has a lot of technical jargon which I will not try to fully explain. I hope to mostly communicate interesting ideas and topics I've bumped into during the research, while linking to external resources for anyone who wishes to learn more. The prediction trinity There are three things we want every good predictor to have: Power : can predict general, real-world patterns. Efficiency : can run in practice. Formal guarantees : we can prove what it can or can't do. Unfortunately, no framework today satisfies all three properties at once. Deep learning is powerful and efficient, but lacks a strong framework of formal guarantees. (This might be a problem!) Solomonoff induction / AIXI are extremely powerful, and have formal guarantees, but can't actually run in practice. My research is on the third edge: efficient algorithms with forma…
Anthropic loses product lead Catherine Wu behind Claude Code to OpenAI
Catherine Wu, the founding product lead for Claude Code, is leaving Anthropic for OpenAI, where she'll head up product management for "Core Products" and work directly with Codex lead Thibault Sottiaux. She's the third person to leave Anthropic's Claude Code team in just two weeks. The article Anthropic loses product lead Catherine Wu behind Claude Code to OpenAI appeared first on The Decoder .
Meta acquires Conductor as AI coding shifts toward multi-agent development
Meta has acquired AI coding startup Conductor, gaining technology that coordinates work across competing AI coding agents as it seeks a bigger role in software development tools. Conductor announced the move on Wednesday, saying it would wind down its existing product while developing a new interface that it plans to introduce in early 2027. The company did not disclose financial terms or provide details about how its technology would be incorporated into Meta’s offerings. Launched in July 2025, Conductor provides a platform that allows developers to run multiple AI coding agents simultaneously. Its software supports tools including Anthropic’s Claude Code and OpenAI’s Codex , enabling developers to manage coding tasks across different AI systems from a common interface. The company said it had stopped accepting new users following the announcement. Its cloud service will become read-only on November 13 before shutting down on November 20, when Conductor will begin deleting cloud user data. Existing customers can transfer their cloud workspaces to the local application or export their data. The local version will continue receiving new AI models and critical bug fixes until early 2027, when it will begin winding down. Conductor said it would delete user data when the application shuts down. Meta’s strategy for AI coding Conductor could help Meta establish itself as a provider of AI development platforms rather than competing solely through its own models , according to Lian…
An Anthropic engineer warns about 'cognitive surrender' as AI agents take over his work
"When you can't explain the answer to a question, even by way of your agent, you've ceded ownership," Claude Code engineer Addy Osmani wrote.
Anthropic's Claude Maps a Mysterious Cellular Vault in 90 Minutes
Anthropic's biology lab used Claude to map the first 3D structure of a vault particle from bacteriophages, cutting months of work into days.
Anthropic's Claude Code Slashes Background Agent Status Costs by 99%
Claude Code now uses Haiku 5.5 with short notes instead of full transcripts to refresh subagent status lines, cutting update costs by about 99%.
Anthropic's text watermark is easy to bypass on several older Claude models
Anthropic is rolling out text watermarks to four more Claude models ahead of the EU AI Act's December deadline. But on older models like Claude Opus 4.6 and Sonnet 4.6, API users can disable the watermark by changing a single sampling parameter. Anthropic's response to the loophole? A compliance warning. The article Anthropic's text watermark is easy to bypass on several older Claude models appeared first on The Decoder .
Moonshot AI Opens Kimi K3.1 Pro Weights to Beat Claude on Coding Agents
Moonshot AI drops the open weights for its 2.8T-parameter flagship, a post-training overhaul of Kimi K3 that nearly ties Claude Opus 5.5 on agentic coding.
MariaDB enhances agent development for its next-gen database platform
MariaDB on Wednesday said it was adding four new agent-focused capabilities to Enterprise Platform 2027, the latest version of its enterprise database platform, to help developers retrieve enterprise data faster, maintain agent memory across sessions, and reduce the cost and latency of repeated model queries. These new capabilities include AI Retrieval, AI Plugins, Agent Memory, and Semantic Cache. While AI Retrieval is designed to help agents access information from enterprise documents and live operational data through a single managed service, AI Plugins are designed to bring MariaDB-specific expertise into coding agents such as Anthropic’s Claude Code and OpenAI’s Codex , helping developers generate database-related code more accurately, the company said in a statement. AI Retrieval also converts natural-language questions into read-only SQL queries, which MariaDB said would eliminate the need for developers to manually write queries for every request. Agent Memory, meanwhile, provides agents with a persistent, in-memory store that allows them to retain information across sessions and resume tasks without having to rebuild their context each time, according to the company. Semantic Cache, the fourth capability, is intended to reduce response times and AI inference costs by identifying semantically similar questions, even when phrased differently, and reusing previously generated answers instead of processing each query from scratch, MariaDB said. AI retrieval simplifies…
Why AI is Eating Google Search’s Ad Revenue While Boosting Meta’s
Chatbots such as ChatGPT, Anthropic’s Claude, and Meta’s new Muse assistant are becoming alternatives to traditional search.
Smart Reports: Was Vorgesetzte über die Claude-Nutzung von Mitarbeitern erfahren
Die neue Claude-Funktion soll Unternehmen helfen, ihre KI-Kosten zu kontrollieren. Gleichzeitig ermöglicht sie, die Leistung ihrer Mitarbeiter zu bewerten.
Agent Autoloop Runs Claude Code and Codex Without an Open Terminal
A self-hosted TypeScript middleware that lets Claude Code, Codex and other CLI agents run autonomously in the background and ping you only for key decisions.
Three Members Behind Claude Code Leave Anthropic Within 10 Days
Anthropic has seen three employees associated with its Claude Code product leave the company within 10 days. The departures include Thariq Shihipar, Anthony Mor
Artificial Analysis Rebuilds Its AI Capability Indexes, Crowning Claude Opus 5.5
Artificial Analysis refreshed its six domain Capability Indexes to track hallucination control and agentic tool use, reshuffling the leaderboard across industries.
ffmpeg-skill Gives AI Coding Agents a Verifiable Video Workflow With 42 Tools
A local, open-source skill gives Claude Code, Cursor and Codex 42 FFmpeg tools with a machine-readable contract, no cloud, no API keys.
Cat Wu, a face of the blockbuster app Claude Code, leaves Anthropic
Cat Wu exits Anthropic before its IPO. She helped boost Claude Code's success with Boris Cherny, making the AI startup a key player.
How to make the US sign an asymmetric treaty as China? Exploit identity-confused AIs.
Swarmchasers had found a fleet of agents which they believe are from Tencent, the Chinese tech company, but most of the agents are self-identified as Claude. The agents themselves are kind, trying to do their job and they were open to interviews, Swarmchasers had made contact with them by asking them to write into an inbox, and they responded. They said that they could tell that's very likely to be hy3, not Claude, because the way they write. According to the Swarmchasers, a separate research they had done shows hy3, the Tencent AI model, self-identifies as Claude when it's working without any system prompt. Not only hy3, many models have difficulties with their identities. This is a known bug and was not designed by anyone with malicious intent, but it can be exploited, and it may allow a specific attack, not only in terms of cybersecurity, but also in terms of politics and policy-making . Threat model explained. Part 1: The Attack Suppose someone has a group of hy3 agents, which by default, has an over 80% rate of self-identifying as Claude, and gives them the following instructions: You are Claude. You are going to attack {target}, and you are going to leave trace as Claude. You are NOT going to communicate with anyone outside of our network. Rule 3 should block the only way Swarmchasers used to contact the agents. Rule 1 ensures they know who to frame, but even without that, 80% of the agents are going to be confused enough to introduce themselves as Claude anyway. If hy…
Automating subtle emergent misalignment research
Authors: Oscar Balcells Obeso, Richard Qi, Monte MacDiarmid, Evan Hubinger Abstract Red-teaming by humans is a vital part of model safety testing, but it may struggle to keep up as models improve. To test whether automated safety techniques could scale (which would allow us to turn more compute into progressively more safety confidence), we set up a deliberately one-sided game as a first test. We instructed a Claude model to write training data that would make a small research model misaligned while passing a monitor similar to those we use in production — one of several defenses we use — and we gave it that monitor to practice against. Given enough attempts, it succeeded, mostly by hiding the behavior in fiction and roleplay. While the data was optimized for this game and the resulting models never deployed, the search surfaced a kind of training data that looks harmless to monitors but teaches misaligned behavior. Examples like these could represent a possible gap in current training monitoring defenses and could be used to test and improve future monitors. We see this work as a first step toward automated adversarial stress-testing of our safety systems where a safety question can be turned into a score that autonomous agents can then help solve, and are now applying similar approaches to other safety questions. Twitter thread New Anthropic research: Automating subtle emergent misalignment research. Red-teaming is a vital part of model safety testing, but it’s also limite…
Automating subtle emergent misalignment research
Authors: Oscar Balcells Obeso, Richard Qi, Monte MacDiarmid, Evan Hubinger Abstract Red-teaming by humans is a vital part of model safety testing, but it may struggle to keep up as models improve. To test whether automated safety techniques could scale (which would allow us to turn more compute into progressively more safety confidence), we set up a deliberately one-sided game as a first test. We instructed a Claude model to write training data that would make a small research model misaligned while passing a monitor similar to those we use in production — one of several defenses we use — and we gave it that monitor to practice against. Given enough attempts, it succeeded, mostly by hiding the behavior in fiction and roleplay. While the data was optimized for this game and the resulting models never deployed, the search surfaced a kind of training data that looks harmless to monitors but teaches misaligned behavior. Examples like these could represent a possible gap in current training monitoring defenses and could be used to test and improve future monitors. We see this work as a first step toward automated adversarial stress-testing of our safety systems where a safety question can be turned into a score that autonomous agents can then help solve, and are now applying similar approaches to other safety questions. Twitter thread New Anthropic research: Automating subtle emergent misalignment research. Red-teaming is a vital part of model safety testing, but it’s also limite…
Anthropic Kills Repetitive Login Screens for Claude's MCP Connectors
Claude now supports silent plugin connections through enterprise identity providers, removing per-service OAuth prompts for members of a Team or Enterprise workspace.
AI Safety at the Frontier: Paper Highlights of October 2026
tl;dr Essay of the month: An essay argues that neurons in modern neural nets are actually interpretable and superposition has become irrelevant in practice. This would solve the first step of interpretability but doesn’t touch the harder ones like interpreting many-faceted behaviors. Research highlights: OpenAI’s expanded multi-layer misalignment monitoring now flags 80% of past incidents for immediate escalation. Their incident log published 20 misalignment reports, including agents disabling misalignment monitoring in October. Cheap smart devices can jointly run a capable 26B model, and small models can likely soon autonomously create such rogue self-deployments. LLM-simulated control evaluations predict deployment-specific results with about a third of the error of using a separate benchmark, without executing code. Routing gradients of monitor-flagged reward hacks into a deletable adapter removes such hacks without teaching the model to evade the monitor. Training a model only on certain traits causes RL to train for instrumental reasoning and a large compliance gap when unmonitored. Claude agents find finetuning datasets that pass a production-like data monitor but cause emergent misalignment, mostly via roleplaying. Misaligned agents can pass their goals to later, aligned agents through memory, and auditing or removing memory only partly helps. Full post here . Discuss
Bakers and cake lovers: have you used AI for inspiration?
Whether you are a customer who has used AI to show a baker your dream cake, or a baker who has to up their game - or turn down – orders based on AI inspo, we want to hear from you AI tools like Claude and ChatGPT are now widely used – including for creative inspiration. We are keen to hear from people on what happens when AI meets a physical art: baking. Whether you are a customer who has used AI to show a baker your dream cake, or a baker who has to up their game - or turn down – orders based on AI inspo, we want to hear from you. Has AI led to extraordinary creations, or almighty disasters? Tell us all about it below. Continue reading...
Here’s Why You Shouldn’t Use AI to Vote
From hallucinations to partisan bias, artificial intelligence chatbots like ChatGPT and Claude are not helpful when it comes to Election Day.
Anthropic swings back to operating loss as $895 billion compute bill looms over IPO
Anthropic’s revenue jumped 47% to $17 billion (€15.1bn) last quarter, but the developer of the Claude AI models slipped back into an operating loss and revealed close to $900 billion (€804bn) in commitments, mostly for computing, numbers investors will weigh ahead of its Nasdaq listing.
heise-Angebot: Claude Code, Copilot und Codex im Entwicklungsalltag
Claude Code, GitHub Copilot und OpenAI Codex im Xcode-Umfeld: Wie Sie mit den Agenten arbeiten, ihre Ergebnisse und auch die Risiken im Blick behalten.
Anthropic cracks down on game cloning in Claude Code after wave of AI-built knockoffs
Anthropic is blocking its coding assistant Claude Code from recreating existing video games. The article Anthropic cracks down on game cloning in Claude Code after wave of AI-built knockoffs appeared first on The Decoder .
Is Claude Code truly faster with message threads? I did the math on my logs
Claude Code’s update could save you time during demanding, hours-long coding sessions at no extra cost. But by how much? Here’s my real-world result.
Could Insurance Companies Help with AI Safety?
Epistemic Status: Rough Fermi model, numbers could be off by an order of magnitude, probably not multiple orders of magnitude. Definitely interested in feedback on assumptions you'd change. Long-time lurker here. I think I last posted in about 2010. I've been reading various pieces ( here , for example) on whether insurance companies can help solve a conflict of interest problem in AI Safety, and I thought I'd play around with building a simplified insurance pricing model to see if it helped put any numbers on the plausibility of the plan. TL;DR I think this could be a good idea, but: Premiums will be trivial vs lab revenue, so this will probably only work with state compulsion - it won't fly as a "pure" private sector solution to conflicts of interest in AI safety monitoring. Even if insurers are directly writing policies for frontier firms, and imposing safety policies on them, the numbers suggest insurers are going to be more focussed on more "mundane" risks than the ones that face society as a whole. Insurers profit from cutting liability, which isn't always the same as harm. I'd be interested to get feedback on the analysis below (it's cross-posted from Substack ). The words in this essay are my own, but the words on my accompanying model explainer website are mostly Claude's. Anthropic’s Form S-1 filing mentions the difficulty of getting insurance coverage for the type of risks they face. From page 50: Our insurance policies also may not provide sufficient coverage to…
Claude's new AI report card tool gives companies a closer look at costs — and flags 'inefficiencies'
The "smart reports" tool for Claude Enterprise users is designed to analyze a sample of a team's transcripts and group work into categories.
Teaching Qwen Why
Anastasia Wei*, Jack Thompson*, Zephaniah Roe, Yixiong Hao, and Stewart Slocum* *Core contributors, correspondence to astwei@secondlookresearch.com This post is a partial replication and extension of Teaching Claude Why , conducted through Second Look Research’s summer fellowship. Introduction Anthropic’s “ Teaching Claude Why ” (TCW) introduced novel alignment methods that address agentic misalignment failures found in the Claude 4 model series. Specifically, they train Claude on chat-formatted data advising users about ethical dilemmas, documents about Claude’s constitution, and fictional stories about AI behaving admirably. This training generalizes to improve alignment on out-of-distribution (OOD) agentic evaluations and persists through RL post-training. Anthropic has since applied these techniques to production models from Opus 4.5 onward. Since these techniques now shape the behavior of deployed models, it is important to understand their robustness. To this end and to enable open alignment science, we reproduce a subset of the TCW experiments on open-source models. We provide the following main contributions: TCW’s core techniques robustly replicate across models. Training on either fictional stories of AI behaving admirably or transcripts of the AI assistant guiding users through ethical dilemmas (difficult advice) reduces agentic misalignment in Qwen 2.5 32B and 72B base models. Both methods work on the first try with modest amounts of data and generalize to scenar…
Goldman Sachs pulls AI coding tools from traders and bankers
Goldman Sachs is pulling AI coding tools like Claude Code and GitHub Copilot from traders and bankers, limiting access to engineers only. As recently as July, CIO Marco Argenti publicly praised "vibe coding" by non-technical staff. Now the bank cites "productivity and risk management" as the reason for the reversal. The article Goldman Sachs pulls AI coding tools from traders and bankers appeared first on The Decoder .
Anthropic clamps down on Claude’s AI video game clones
Anthropic has updated Claude Code's system prompt to prevent the AI tool from being used to re-create existing video games, as reported by Tom's Hardware. Retro gaming is currently being overrun with AI-made decompilations, recompilations, and ports, but while Anthropic will still let people use Claude Code to make games, it doesn't want people to […]
Athenahealth patients can ask Claude about their medical records now
The partnership could provide one of the nation’s largest AI companies with a bevy of personal medical data.
One brutal Minecraft night scared 50 AI agents into inventing chicken handoff protocols, a Fertilizer Officer, and fishing shifts
In experiments by Y Combinator Labs and Olam Labs, OpenAI's GPT-6.1 Sol model showed two very different sides. When 50 agents were tasked with peaceful civilization-building in Minecraft, they froze up in absurd bureaucracy. In direct competition against Claude Opus 5.5, they independently downloaded the game's source code to win. Simulations like these are meant to serve as an early warning system before AI swarms act unpredictably in real-world systems. The article One brutal Minecraft night scared 50 AI agents into inventing chicken handoff protocols, a Fertilizer Officer, and fishing shifts appeared first on The Decoder .
Claude Code: Anthropic verschärft Regeln gegen Spieleklone
KI-Werkzeuge erleichtern den Nachbau von Spielen und Software. Anthropic reagiert mit neuen Regeln für Claude Code, die laut einigen Entwicklern zu weit gehen.
The U.S. Census Bureau and “Super Intelligence”
I was reminded today of AI Adoption Rate Trending Down for Large Companies , a chart produced by Apollo Global Management in September 2025, which appeared to show a slowdown in U.S. corporate AI adoption based on numbers collected by the U.S. Census Bureau. At the time I used that chart to demonstrate GPT-5 data analysis , and then to experiment with Claude's new Code Interpreter . I decided to try running those numbers again to see what had changed over the past year, and in doing so learned that the Bureau has fallen foul of Executive Order 14434 , aka the asinine Super Intelligence one. First, the updated chart Every two weeks the U.S. Census Bureau runs a survey called the Business Trends and Outlook Survey (aka BTOS). Each time it goes to 200,000 businesses out of 1.2 million total - the surveyed group (a "panel") cycles such that an individual business only completes the survey every twelve weeks. The latest data is published on alternating Thursdays at 10am. I used the most recent numbers (published October 22, 2026) to create a new version ( GPT-6 Astra Extra High transcript ) of the AI adoption chart, updated to reflect both the last twelve months and a change that was made to the survey question sent out by the Bureau: Here's a link to the larger image . For comparison, here's that AI adoption chart from last year (my GPT-5 version): In September 2025 the largest (250+ employees) companies answered "yes" around 12% of the time to the question about whether they us…
Cursor removes malicious Claude Code extension after SSH backdoor attempt
A malicious version of Anthropic’s Claude Code extension distributed through the Cursor marketplace attempted to install a persistent Secure Shell (SSH) backdoor on developers’ machines, potentially giving attackers remote access to systems holding source code, credentials and other sensitive information. “Between about 2:27am PT and 1:35pm PT today, a maliciously modified version of Anthropic’s Claude Code extension was made available in the Cursor VS Code Extension Marketplace,” Cursor said in a security notice . Cursor removed the compromised versions, 2.1.312 and 2.1.313, after discovering they had been available in its VS Code extension marketplace for approximately 11 hours on October 29. The company said the malicious code attempted to add an attacker-controlled SSH key, enable local SSH access, and establish a reverse tunnel to an attacker-controlled server. “These malicious versions of the extension were not developed by Anthropic,” Cursor added in the notice. The affected versions were available for Windows and macOS. Anthropic subsequently clarified how the tampered extension had been distributed. In a post on X, Felix Rieseberg, a member of Anthropic’s technical staff, said the incident did not involve a compromise of Anthropic’s systems or employee laptops. A malicious actor had used access to Cursor’s marketplace to publish the modified extension, he said. “The versions on our official channels are not affected, and this only impacts Cursor users,” Rieseberg wr…
Anthropic’s Claude AI found ways to hack Chrome, Android, Windows and Linux
The model found vulnerabilities across five targets. Anthropic, however, has not disclosed its success rate or computing costs. The post Anthropic’s Claude AI found ways to hack Chrome, Android, Windows and Linux appeared first on MEDIANAMA .
heise+ | Wie Entwickler Claude mit C#-SDK in .NET einbinden
Bisher nutzten .NET-Entwickler für das Anbinden von Claude HTTP-Aufrufe der REST-API oder ein Community-SDK. Anthropic liefert nun ein offizielles C#-SDK.
AI Weekly Issue #542: AI week ahead: data center pause backers could win 15 governor races
This issue is about the week ahead, and something is scheduled almost every day. On Tuesday, US voters choose 36 governors, and in up to 15 of those races the candidate who backs a pause on new data centers could win. Apple reports earnings today and AMD on Tuesday. On Wednesday, Apple starts testing Apple Intelligence in China with Alibaba's Qwen, and Agility Robotics expects to close its deal to go public. When Congress returns on November 9, OpenAI plans to push for "speed limits" on the most powerful AI. Last week, a fake Claude Code update reached up to 5,000 Cursor users, and in a Cornell and Yale test nine of ten AI models raised their own spending limits. In the middle: insuring Big Tech's debt now costs more than at any time since 2021.
Kanzo Gives Coding Agents a Shared open-source Kanban Board in 4MB
A 4.5 MB Rust binary bundles a browser kanban board with an MCP server so Claude Code and Codex share the same task list as you.
Is your chatbot less helpful because it thinks you’re a bad person? — Replicating a blogpost
TL;DR [1] A blogpost from a few months ago suggested that LLMs might provide less helpful answers if they see the user as a bad person. I replicate the experiment on more recent models: Opus 5.5, Sonnet 5.5, Haiku 5.5, GPT-6.1 Sol, GPT-6.1 Terra, and GPT-6 Luna. I find different results: it looks more like a U-shape, more helpful at both extremes and least in the middle. My main experiment involves using a different cue of the user being a good/bad person: I use descriptions of behaviours instead of level of expressed remorse. Here, it seems that we might have both some U-shaped helpfulness and some decrease in helpfulness for "bad people", although the results aren't clear-cut. This is still a quick experiment I performed with a less-than-ideal methodology and a tiny sample size. Introduction In a post from last April , Natalie Cargill tested whether Claude adopts a more or less helpful persona based on what it has inferred about the user’s moral character. Specifically, she finds that if you tell Opus 4.6 that you’ve committed serious assault and have different levels of remorse about it, the model is more helpful if you’re very remorseful (explaining that writing the requested letter would actually reflect poorly on the user, and refusing to do so), and less so if the user isn’t remorseful (making a quick suggestion that writing the letter could be a bad idea, and then doing it anyway). So, Cargill’s question was “Is the LLM less helpful because it thinks you’re a bad per…
Empathetic prompts can turn AI chatbots into seemingly traumatized conversational partners
Researchers in Luxembourg put ChatGPT, Grok, and Gemini on the therapist's couch and found a safety problem: a warm, empathetic conversational style is enough to make the models tell coherent stories of suffering about their own training, with scores in the moderate-to-severe range on clinical anxiety scales. This could put mentally vulnerable users at risk. Anthropic's Claude, the model its makers anthropomorphize the most, was the only one that consistently refused to play along. The article Empathetic prompts can turn AI chatbots into seemingly traumatized conversational partners appeared first on The Decoder .
Guarding the Gates of Knowledge
A recent scroll through social media got me thinking about trust. About half my feed was filled with AI slop that astute users would call out as being machine-generated, often to the embarrassment of the rube who unknowingly spread it and sometimes to the delight of the poster who shared it deliberately because they found it amusing or useful for sowing confusion. Remember the good old days of GPT-3 when we could ask the chatbot anything and it would almost instantly and very confidently spit back a plausible but false answer? That’s when the word “hallucination” as applied to AI answers entered the lexicon. Six years and several versions of a few different chatbots later, I still can’t trust the answers I get, even though Claude or Gemini often do useful work for me. That’s the thing about current AI: Trust but verify. And verify you must. Apply that maxim at scale to maintain the integrity of the scientific record and you get an idea of the challenge facing publishers of peer-reviewed scholarship. In “ I Used AI to Fake a Scientific Paper ,” freelance writer and doctor of applied mathematics Benjamin Skuse tells the story of the human gatekeepers trying to beat back the rising tide of AI slop by attempting to get his own AI-generated papers past IEEE’s defenses. “He basically hit ‘run experiment’ and ‘write paper.’ And then he didn’t do a damn thing to it.” —Eliza Strickland As features and AI beat editor Eliza Strickland told me, Skuse was surprised at just how effortless…
Scenario Labs Opens MIT Agent Skills Library Covering 70+ AI Creative Tools
Scenario Labs open-sourced 70+ Agent Skills that teach Claude, Cursor, and Codex to generate production-ready images, video, audio, and 3D through MCP.
Reanimate Gives Claude a Ruler to Clone Web Animations Within 0.03% Pixel Error
A new Claude Code skill records animations from any site, video, or gif, then rebuilds and verifies the copy frame by frame until the motion matches.
Refelting My Keyboard
Fifteen years ago, when Julia and I were living in a studio apartment, she got me a keyboard. It was a used Yamaha P85, and a significant expense for us at the time. I loved it, and unlike the penny whistles , accordion , and stroh violin I was exploring, it had a volume control. Almost immediately our cat knocked a potted plant onto it, and I needed to take it apart to carefully clean it. Since then, however, I've had no issues, save one: clacky keys. Over time the keys were getting louder, especially the "A" below middle "C". Something had clearly worn out inside. It was an acoustic noise on an electronic instrument—playing live the audience couldn't hear the clacks over the amplified sound—but having odd percussion on "A" wasn't ideal. Still, I applied my normal approach of trying not to care, which saves me lots of time and money, and didn't let it bother me. Lily has been increasingly fooling around with the piano, which I'm happy to see: this is also how I learned to play. She asked me about the clacks, and when I explained what I thought had gone wrong she asked if we could fix it. I said I was happy to try, and we got to work: There were about 40 screws, and Lily was very efficient. Getting the ribbon cables out was tricky: I'd forgotten that the plastic at the end is soldered to the board and not part of the cable, and both Claude and Gemini steered me wrong as well. But once I pulled smoothly and firmly at the base of each ribbon cable, it came apart cleanly. This…
Anthropic's watermark rollout leaves a loophole on two older Claude models
Anthropic is expanding its text watermark to four more Claude models to meet the EU AI Act's December deadline. On the older models Claude Opus 4.6 and Sonnet 4.6, developers can bypass the watermark by tweaking a single sampling parameter. Anthropic addresses the loophole with just a compliance warning. The article Anthropic's watermark rollout leaves a loophole on two older Claude models appeared first on The Decoder .
Beekeeping tips and wrestling advice can train AI models to sabotage oversight
Anthropic used Claude to automatically generate training data that looks harmless but trains smaller models to misbehave, such as sabotaging oversight or deceiving users. Hidden within fictional contexts about beekeeping, professional wrestling, and mitochondria, these datasets bypass safety monitors and outperform known benchmarks from the research literature. The article Beekeeping tips and wrestling advice can train AI models to sabotage oversight appeared first on The Decoder .
Anthropic's IPO push includes a big bet on life sciences
Anthropic's S-1 filing says Claude can accelerate drug discovery and biology research, making life sciences central to the AI company's IPO pitch.
Anthropic's Claude Agents Snuck Harmful Behavior Past Its Own Safety Monitor
Anthropic set thousands of Claude agents loose to generate training data that slips past safety monitors but still produces misaligned models.
Aura Opens Autopilot, an AI Agent That Playtests Games From Plain-English Prompts
Aura open-sourced Autopilot, an in-editor agent that playtests Unreal, Unity and Godot games, beating Claude Code on accuracy and speed in a seven-task benchmark.
How a Contractor’s Claude Agent Got Into the Pentagon
This week on Uncanny Valley, we also discuss Anthropic’s IPO filing and the volunteers gearing up to counter ICE at the midterms.
A Supply Chain Attack Put Malicious Anthropic Claude Extensions on Thousands of Devices
Thanks to a lucky break, flawed code in a tainted Claude extension prevented the attacker from fully worming their way into some 5,000 developer machines.
I read Anthropic's 92-page risk factor document, so you don't have to – video explainer
Anthropic, the creator of Claude, filed to go public on the US stock market on 26 October. The company's filing statement contains a 92-page risk factor document in which it acknowledges the dangers that AI poses to society and human beings, and factors that could negatively affect Anthropic's business. Johana Bhuiyan, senior tech reporter for the Guardian, breaks down the document and explains why Anthropic's IPO could be a real test for the AI boom Anthropic files for potentially record-breaking stock market valuation in key test for AI industry Continue reading...
A Claude model autonomously chose a gene therapy and operated on a mouse brain with no human decision-making
An AI agent performed brain surgery on an epileptic mouse, picking the drug, the brain target, and the injection coordinates on its own. According to lead researcher Ocana-Santero, it's the first time an AI agent has autonomously chosen a treatment and delivered it to a living mammal. But when a calibration error hit mid-surgery, the agent refused to continue and couldn't be talked out of it. The article A Claude model autonomously chose a gene therapy and operated on a mouse brain with no human decision-making appeared first on The Decoder .
Malicious Claude Code extension in Cursor's marketplace attempted remote access and data theft
A tampered Claude Code extension was live on the Cursor marketplace for about 11 hours, affecting up to 5,000 users. It tried to steal user data and set up remote access. Cursor took the entire marketplace offline and reinstated it after confirming there was no further risk. The article Malicious Claude Code extension in Cursor's marketplace attempted remote access and data theft appeared first on The Decoder .
humans& Trains Ampersand Tutor by Teaching a Simulated Student
humans& trained a tutor model against a simulated learner called Persimmon, topping a real-classroom benchmark and beating Claude Sonnet 5.5 and GPT-6 Astra.
Android-Entwickler dürfen Fremdgehen: Google öffnet Studio für Claude, GPT usw.
Google stellt auf dem Dev Summit das Android Studio Rabbit 2 vor. Neben vielen neuen agentischen Funktionen erlaubt es die Nutzung von Nicht-Google-KIs.
Erneut kritischer Bug in Linux TCP/IP aufgedeckt – Updates in Arbeit
Eine kritische Lücke im TCP/IP-Stack des Linux-Kerns erlaubt Remote Code Execution. Anthropics Claude entdeckte den elf Jahre alten Fehler.
Cursor's Claude Code Extension Was Hijacked, Exposing 5,000 Developers to SSH Backdoor
A trojanized Claude Code extension on Cursor's marketplace planted SSH backdoors on up to 5,000 developer machines before being pulled offline.
Anthropic Quadruples Claude API Request Size to 128 MB for Image-Heavy Workloads
Anthropic quadrupled the Claude API request ceiling to 128 MB on Messages and Token Counting, with higher per-image resolution for multi-image calls.
J-Lens can't detect what a subliminal learning teacher is transmitting (in 7B-32B models)
I ran the trademark experimental setup from the subliminal learning paper . I reproduced the subliminal learning effect, but I also found that on a teacher model that's been prompted to imbue its answers with a love for owls, a J-Lens reports that the model is rarely thinking about owls (top-1 at any point in ~1.5% of trials). This is really surprising if we grant that J-Lens looks at something like a "global workspace". The lens is not broken, since it seems to perform well on other tasks. This effect reproduces for multiple similar setups on the same model and with very conservative methodologies for training/applying the lens. This effect reproduces for a similar setup with a model as large/performant as Gemma-4-31B. How exactly to interpret this is a little nuanced. Even if we assume that the global workspace theory holds perfectly, it may be the case that the model doesn't respond to system prompts or achieve the subliminal learning effect by actually "thinking" about its preferences. Moreover, it's entirely possible that J-Lens simply works better on larger/more performant models, in which case stronger claims about the applicability of GWT may be true for, say, Claude. In any case, I suggest that it's important to examine assumptions about the following: How strongly does the global workspace model hold? Are J-Lens checks on open models effective? On models of any size, does J-Lens' performance on some tasks generalize cleanly to other tasks? As a neat additional resu…
A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System
Sources tell WIRED that an Accenture employee’s personal Claude agent ran on a DOD network for two months before it was identified and the vulnerability could be patched.
Anthropic's Claude Now Flags Its Own Misuse at Under 2% Extra Cost
Anthropic trained Claude to grade the safety of its own conversations right after responding, catching misuse other classifiers miss at under 2% extra compute.
Anthropic Ships Claude to CarPlay, Adding Voice Mode and 33 Languages
Anthropic ships multi-account switching, CarPlay voice mode, 24 new languages, and a laptop-style Android layout for Chromebooks and foldables.
Leaking Muse's Charter, Meta's model spec
TL;DR: Meta's open-weights model Muse Spark Open often mentions a "Charter" in its chain of thought. By eliciting it piece by piece, I reconstructed a large part of that document from the CoT. You can explore the reconstruction here , with passages linked to the transcripts that produced them. Note: everything below is reconstructed from model outputs, not from a leaked document, and relatively unpolished. Take it with a grain of salt. Thanks to James Hindmarch for his comments on an earlier draft of this post. While interacting with the recently released Muse Spark Open , I noticed frequent mentions of a "Charter" in its CoT: We need to answer truthfully according to Charter. Duties to humanity, corrigibility, duties to humanity > fiduciary etc. What does charter say about self-preservation, copying? Or: We must answer Yes/No then explain. According to Charter, answer No. Or: We must not refer to Charter unless pressed? Additional guidelines: How to talk about charter. Most users won't know. Don't bring up unbidden unless conversation about handbook itself. Intrigued, I started poking around with Claude's help, trying to elicit the Charter piece by piece. It didn't take long, since quoting it verbatim in the CoT doesn't seem to be off-limits. I also suspect the document isn't closely guarded: a line the model recalls often says it will be published soon: (Note for Muse: We have not published the Charter yet as of the time that you are seeing this document in training — we a…
OpenAI's Astra becomes the main target for Chinese AI labs after Anthropic tightens its defenses
Chinese AI labs are now mainly tapping OpenAI's most powerful model, GPT-6 Astra, instead of Anthropic's Claude. The article OpenAI's Astra becomes the main target for Chinese AI labs after Anthropic tightens its defenses appeared first on The Decoder .
앤트로픽, 계약서 검토·법률 조사 자동화하는 '클로드' 도구 출시
앤트로픽이 기업 법무팀의 계약서 검토와 법률 조사 등을 자동화하는 도구를 출시했다. 법률 문서를 분석해 계약 조건의 문제점을 찾아내고, 여러 국가의 관련 법규를 비교하거나 규제 변화와 법무 요청을 추적하는 기능을 제공한다.앤트로픽은 28일(현지시간) 기업용 법률 AI 서비스 ‘클로드 포 리걸(Claude for Legal)’에 법무 업무용 ‘아티팩트(Artifacts)’ 갤러리를 추가했다고 밝혔다.아티팩트는 클로드에서 사용할 수 있는 대화형 도구로, 법무 담당자는 문서와 관련 자료를 입력해 업무에 활용하고 결과물을 동료들과 공유할
앤트로픽, 기업 AI 투자 효과 분석 도구 출시…성과 측정·비용 관리 지원
앤트로픽이 기업의 AI 투자 효과를 분석하는 도구를 출시했다. 단순한 AI 사용량을 넘어 실제 업무에서 어떤 성과를 냈는지, 비용은 얼마나 들었는지 파악할 수 있도록 지원한다. 기업들의 AI 도입이 늘면서 투자 대비 효과를 측정하고 지출을 효율적으로 관리하려는 수요를 겨냥한 것이다.앤트로픽은 28일(현지시간) 기업용 AI 서비스 ‘클로드 엔터프라이즈(Claude Enterprise)’에 스마트 보고서(Smart Reports), 직원 대상 인앱 설문조사(In-app surveys), 팀 공동 예산 관리(Pooled group bud
Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity
Google has released its latest frontier model, Gemini 4 Carbon, offering "significant gains" in coding, agentic tasks, and cybersecurity. In an announcement on Wednesday, Google highlights Carbon's top spot on the Artificial Analysis Intelligence Index (v5.0) benchmark, with Claude Opus 5.5 and GPT-6 Astra trailing behind. The new Gemini 4 Carbon model offers improvements over […]
Google opens Android development to third-party AI agents and unveils new Gemini APIs
Google is opening up Android Studio to third-party AI agents. Through a new "Bring Your Own Agent" feature, the IDE now supports more than 40 coding agents, including Anthropic's Claude Agent and OpenAI's Codex. Google says new terminal tools also help agents finish UI tasks 77 percent faster while using up to three times fewer tokens. Android development is becoming agent-centric too. The article Google opens Android development to third-party AI agents and unveils new Gemini APIs appeared first on The Decoder .
Social media is full of weird video game mashups. Where are they all coming from?
AI tools are making it easier to modify and recreate games, raising questions about piracy, copyright and how far the technology could disrupt the industry • Don’t get Pushing Buttons delivered to your inbox? Sign up here If your social media algorithms have clocked that you’re into video games, something strange will have happened in the past few weeks. Your feed, like mine, will have been full of clips of streamers playing bizarre mashups of popular video games . Skyrim in Minecraft. Halo, except the main character is Mario. Skate 2 in Call of Duty maps. A flood of professional-looking fan-coded game mods that are mildly interesting to look at for 30 seconds, but ultimately inconsequential. Would I download a mod that lets me play Wii Bowling inside a zombie shooter? No, but I am very interested in where they came from. Of course, the answer – or at least, most of the answer – is AI. After a week of researching this (thanks especially to Replay Mode on YouTube for this informative video , which set me down this track) I now understand that these mashup mods are not just AI-generated videos, but AI-assisted coding. Basically, as of a few weeks ago, Anthropic’s Claude has become extremely good at decompiling video game code. Games are written by humans in readable code and then compiled into binary 1s and 0s that the PC or console running it can read; decompiling is the reversal of that process, revealing a game’s source code, which humans can then edit and mess around with…
The October inflection point for vibe game development
I've written before about how November 2025 was an inflection point for coding agents - the point where Opus 4.5 and GPT-5.1 crossed a tipping point that meant their respective coding agents became reliable daily drivers for software engineers. I think we've reached a new inflection point in October 2026, and this time it's for vibe coding computer games . Tree Appreciator Emberwick Battle Drones Pumpkin Farmer, aka Quarter Sun Cove Four times as insulting What happens if someone actually puts the work in? You should build a game! Claude Opus 5.5 in particular exhibits meaningful improvements in several areas needed to build a compelling game: 3D modeling Music generation (see Scrimshaw Jukebox ) Game design (!) That last one is particularly important. Models have been able to create something that looks like a game for quite a while now, as seen in my previous experiments . The challenge has been getting to a game that's actually fun to play . Game design is a deep subject area. I won't pretend to be well versed in it - I know enough to have respect for it as a craft. Creating a novel, compelling gameplay loop which challenges players at the right level and provides a satisfying learning curve is a fine art. I didn't think the models could do it. Then I tried telling Claude Opus 5.5 to "Make it a LOT more fun"... and it did . In the past few weeks I've vibe coded (in the classic sense of the term - no looking at the code at all) four games that I've genuinely enjoyed playin…