AI/ML News & Innovations Hub

US Army Launches Detachment 201: Swears in OpenAI exec as Lt Colonel, next to Palantir

Kevin Weil , no longer with OpenAI, but still self-identifying as LTC, has a new startup rumored. https://www.inc.com/georgia-fearn/kevin-weil-left-open-ai-months-ago-now-seeking-150-million-dollars-for-startup/91389531 Meanwhile, another year, another group of tech industry inductees into the Army. www.army.mil – 10 Jun 26 Army commissions second cohort of tech executives into Executive Innovation... JOINT BASE MYER-HENDERSON HALL, Va. — The U.S. Army commissioned its second cohort of senior technology leaders into the Executive Innovation Corps, kno... Paywalled: Twelve Executives Who Left OpenAI in 2026 - Business Insider

Did OpenAI increased the daily amount for incentivized tier?

@VeitB I checked again, and you were right. I was charged correctly, the bug is in the usage display. The incentivized tier combines two pools, all types of tokens, and with models that are priced very differently, making the charges difficult to reconcile. I overlooked this in my earlier review. So, the bug affects both the usage page and the usage API, not the billing. Thank you all for your help. Unclear usage bothers me more than the actual charges.

Paid ChatGPT Business Workspace Locked Into Free Tier "Monthly Limit" inside Codex

The issue right now is that their support team can just close tickets at will without actually solving the issue.

Problem with tool_choice causing caching issues

There’s no good reason to turn off the “implicit” caching by using the top-level explicit write parameter in a growing chat context, unless you expect to never reuse the API call “write”. That marks the final turn part as a breakpoint automatically. The unexpected symptom, that breaks some use-cases because of the limit of 4 effective write points and 50 total retained (per prompt_cache_key?), is that in “explicit” every input turn that you want to potentially match a previous cache write still has to be marked as a breakpoint, otherwise there is no hit. This is either undocumented or a flaw in the implementation. It may be that the default with no additional parameters is the equivalent of all turns being marked “implicit” with the effect of four potential writes at various lengths. Marking your own four (or more with the oldest ignored for writes) does not cost 4x. It sounds like what you need to do is: not use prompt_cache_options and/or additionally mark the tool return turn before the modified turn as a breakpoint, even when it is in the past, so you can “branch” a cache match from there. use unique prompt_cache_key per chat. As I described before, an oversight (or simply disregard of developers) is that only the Responses API endpoint supports explicit cache breakpoint on tool output return, not Chat Completions and its “tool” role. And functions are blocked anyway unless you use “none” for reasoning effort.

This week’s top stories — 9-15 Aug

Presight Q2 revenue jumps 36%; OpenAI brings inference residency to UAE

ChatGPT Business account did not receive the Codex 15M-user hard reset

@OpenAI_Support An answer for affected Business account users would be appreciated. The language in the reset post was specifically everyone as quoted below. “Old news actually from a bunch of days ago, but crossed that 15M. Enjoy a nice reset everyone. Landing in the next hour or so, go /fast.”

Automatic Codex resets are wasting paid Pro usage

Thanks to you for formating everything

Schedule a new hackathon in Los Angeles area

I see that. I will certainly be around…

AGENTCODI: The Codex Workflow on Android

Hello everyone, I wanted to use Codex properly on Android without first setting up Termux and then maintaining a separate Node/npm environment. What started as the fairly harmless thought, “This should be possible,” has since turned into AGENTCODI and a surprising amount of work. The currently released version is 0.4.5 Early Access and runs on ARM64-v8a devices with Android 10 or newer. AGENTCODI is a standalone Android app for the Codex workflow. It supports authentication through ChatGPT or an OpenAI API key, new and existing Codex threads, live response streaming, and the display of reasoning summaries, plans, commands, file changes, and tool activity. Approval requests can be accepted or rejected directly inside the app. Available models and reasoning levels are loaded dynamically. There is also a dedicated project workspace from which individual files, generated images, or the entire workspace can be exported as a ZIP archive. AGENTCODI’s application and test code is written in Java and C++. The interface currently supports English and German and automatically follows the device language. A surprisingly large part of the development process now consists of finding out how far this kind of workflow can be pushed within Android’s application model. Android occasionally answers that question in its own uniquely charming way. A good example of this is described in my Developer Diary . I spent several weeks putting quite a bit of work into keeping Node.js out of future AGENT…

Develop a chatbot on WhatsApp

Don’t let OpenAI go-to-market salespeople hear about this 5M+ annual billings opportunity. See that the company wants to pay that kind of money for AI inference, $40 bucks a month per-seat for usage likely when there is direct billing of frontier models and encouragement. Find out the goals and the specifications, and survey if AI in general can meet that. Contract with a service provider who understands development and delivering turnkey AI solutions, and who can provide something more reliable and satisfactory than anything relying on even another third-party such as WhatsApp, which will degrade the experience.

Function calling + Assistants API: any pattern for capping cost when the model chooses to call a paid function repeatedly?

On your second question first: raw chat completions (or Responses) is easier, and it isn’t close. You’re the one running the tool loop there, so there is no interception step - you just don’t call the paid thing until you’ve checked the budget. Assistants puts its own run lifecycle in the middle. You can still catch it at requires_action before you submit_tool_outputs, but you’re working inside their orchestration rather than your own. For the confirmation part, don’t cancel the run when it goes over. Cancelling leaves the user with an answer that stops and nothing on screen saying why. We check the budget before executing, and if it’s over we submit a tool output like {“error”: “budget_exceeded”, “spent”: 4.10, “limit”: 5.00} and let the model explain it in its own words. The approval case is the same shape: hold the call, show the user the arguments and what it will cost, then either run it and submit the real result, or submit the refusal. Worth adding since the reply above is about migrating: all of this lives in your code rather than the API, so it carries over to Responses unchanged.

Can OpenAI detect AI-generated images?

I am developing a website to eliminate the AI generate media in our website like other platforms using it so how to use API key of OpenAI in our website?

Provider-independent identity and delegated authority for external AI agents

I’ve been working on a provider-independent governance boundary for external AI agents and would value technical criticism from people building with OpenAI’s agent stack. The premise is simple: Authenticating an artificial agent should establish who is acting—not automatically establish what it is authorized to do. The prototype separates: external-agent identity from authority narrowly scoped and time-bounded delegation deterministic execution/approval boundaries governed persistent state with provenance action receipts linking principal → delegation → agent → workflow → action → result revocation and recovery while preserving the evidence chain The architecture is: External Agent Provider → Governed Agent Adapter → Deterministic Governance Kernel In the current executable demonstration, a simulated external agent begins authenticated but with zero ambient authority. It then receives narrowly bounded capabilities, encounters execution/approval/denial boundaries, produces deterministic evidence, and is finally revoked while the prior evidence chain remains intact. Important limitation: the provider transport is currently deterministic and simulated. I am not claiming a live OpenAI, Codex, Grok, or other provider integration. The goal is not to replace provider-native authorization. It is to investigate what governance should exist around increasingly capable provider-supplied workers when identity, delegated authority, persistent state, execution evidence, and recovery need…

%ToolMatch v1.0.0 pending review since July 30 — request for status check

I encountered the same situation, my plugin (unclebg) and reference id: C-FbcyAa3uc7YW

OpenAI's Computer History turns your clicks and keystrokes into a searchable ChatGPT memory timeline

OpenAI's Computer History records clicks, keystrokes, and app switches on Mac and turns them into a searchable timeline for ChatGPT and Codex. The data is stored locally as unencrypted Markdown files. OpenAI says it's not used for AI training, but memories that feed into chats may still end up as training data. The article OpenAI's Computer History turns your clicks and keystrokes into a searchable ChatGPT memory timeline appeared first on The Decoder .

LifeTime Token Usage Endpoint/Dashboard

Please Please Please, could you provide us with lifetime token usage endpoint and a dashboard? It would be so cool to display it on my profile pages, that I have used this much tokens until now. Even better if it shows highest usage was in this month, my average is this and so on. Since token consumption has become equivalent to git commit histories, lifetime token usage display would be a thing to boast in my profile. Also, when developers see other high performing developers using crazy number of tokens, it will inspire them to use more, thus benefitting OpenAI! I hope someone at OpenAI reads this and makes it happen.

Building agentic workflows with SageMaker AI and Bedrock AgentCore

Learn how to combine OpenAI-compatible endpoints on Amazon SageMaker AI with Amazon Bedrock AgentCore runtime to build a multi-agent workflow where each specialized agent uses the model best suited to its job. This post also shows how to get token-level observability from SageMaker endpoints that Strands Agents does not instrument by default.

Anyone having difficulty getting a response from OpenAI Support?

Hi everyone, Has anyone had difficulty getting an issue resolved by OpenAI Support? My case has been pending for about a week with no response. I have followed up through the support chat and email, but still haven’t received help from a human agent. When I try to open a new ticket, the AI support system tells me to go back to my existing ticket and automatically closes the new one as a duplicate. But when I go back to the original ticket, nobody responds. I’m basically stuck in a loop. Has anyone experienced this and successfully managed to reach a human OpenAI support agent? If so, how did you do it? Thanks for any advice.

1 Million Context to enable professional workloads

I would also add it’s important to look at a problem piece by piece. It’s not “enter the prompt and walk away” yet.

My OpenAI Account Predates ChatGPT — Early Adopter Recognition?

What’s funny historically, I guess, is which model generated those 1 trillion tokens? The Single Value (Wolf) Function and Perspective Collapse Everything compressed into a single value

Codex Weekly Limits Are Draining Way Too Fast — Is This a Bug?

As I’ve stated elsewhere I believe Max and Ultra simply aren’t worth the expense. Sol High for coding tasks is very good indeed. The only criticism I have is Codex isn’t yet capable of doing “everything in one go” imho. You have to tackle things piece by piece. However it’s very good at that. I’m not convinced you can "fire and forget* long tasks reliably either and that modus operandi will land you with too much rework and wasted quota. As far as “shrinking quotas” are concerned I’m working on a tool to make that more obvious if it’s happening at all.

Please Don't Deprecate GPT-Realtime Before a True Replacement Exists

I have also come across similar issues with all newer models. The closest acceptable model after v1 was v1.5. But 2, 2.1 and 2.1 mini have an issue with consistency. The biggest issue preventing me from using the newer models in production is with the use of unnecessary preamble for very short tool calls even with explicit prompting to avoid preambles. Could someone at OpenAI provide assurance that v1 will not be deprecated before an acceptable replacement is provided?

OpenAI I have a problem on DALL·E in GPTs in ChatGPT

Subject: Feature Request: Please add a toggle for Classic / Legacy DALL·E and ChatGPT Styles Message: Hello OpenAI Team, I am writing to ask if you can please bring back the classic, raw, and artistic generation styles from 2022–2024 as an optional setting. The modern ChatGPT and DALL·E updates are very powerful, but the image rendering feels way too corporate, waxy, and hyper-polished. The original, classic styles had an organic texture, unique artistic flaws, and a raw digital aesthetic that made them incredibly special. Please consider adding a “Classic Mode” or a “Legacy Toggle” in the settings for free users, so we can choose to generate text and images (to be ChatGPT popular and I will use ChatGPT usefully and many days) using the classic unpolished styles instead of forcing the modern, smooth digital look. Thank you!

Critical ChatGPT Business Issue: Workspace Invitations Are Failing for Most of Our Users

Yeah, OpenAI support is useless. I wanted to switch from Claude to ChatGPT because it’s cheaper, but I think I’ll go back.

GPT-5.6 Sol goes 14x faster as OpenAI launches Ultrafast mode powered by Cerebras

OpenAI is launching "Ultrafast," a new inference mode that delivers GPT-5.6 Sol at up to 750 output tokens per second, powered by Cerebras hardware from their $10 billion partnership. Together with "Standard" and "Fast," Ultrafast creates a three-tier pricing structure that turns inference speed into its own product. The article GPT-5.6 Sol goes 14x faster as OpenAI launches Ultrafast mode powered by Cerebras appeared first on The Decoder .

OpenAI loses its AI ethics lead

OpenAI has lost its AI ethics lead Chloé Bakalar just a year after she joined the company, the Financial Times reported . Bakalar has maintained a silence and has yet to update her LinkedIn profile , but if her departure is confirmed then it will add to the list of OpenAI executives who have quit in recent months. Other departures include robotics chief Caitlin Kalinowski , who left the company over its deal with the US Department of Defense; researcher Zoe Hitzig, who quit in a very public way by writing an article in the New York Times; and Johannes Heidecke, head of safety systems. OpenAI’s ethical stance has been called into question following an attack by OpenAI models on Hugging Face . The departure of its sole ethicist will add to the pressure on the company. In her year at OpenAI Bakalar focused on ethical approaches to model development, looking at how humans interact with AI and examining machine consciousness, according to the FT report. Bakalar had considerable expertise in the area. She was previously at Meta, where she developed the company’s ethics programs, but has also held several positions at prestigious universities on both sides of the Atlantic. Her departure will cause some anxiety at OpenAI as it continues to prepare the ground for its IPO . This article first appeared on Computerworld .

SpaceX Closes $60B Cursor Deal to Challenge Anthropic and OpenAI

SpaceX closes its $60B all-stock acquisition of Cursor, folding the AI coding giant into SpaceXAI to challenge Anthropic and OpenAI in the developer tools market.

Chat GPT pricing strategy

The Bodez ChatGPT Pricing Model Executive Proposal to OpenAI Product & Pricing Proposal: Introduce flexible, compute-based pricing between the Free and Plus tiers. The problem A significant group of users may value ChatGPT highly but use it intermittently. A fixed monthly subscription can be difficult to justify when usage varies dramatically from month to month. At the same time, simple message counts don’t accurately represent OpenAI’s underlying cost: a short factual question and a lengthy reasoning task can require vastly different amounts of compute. Proposed solution Create usage-based tiers with daily compute allowances, rather than relying primarily on message limits. For example: - Free — minimal daily compute - Casual — low-cost part-time tier - Plus — regular users - Heavy/Pro — high-compute users Compute consumption would vary according to the complexity of the request. Intelligent throttling When a user reaches their daily allowance: 1. Continue with a lower-compute model, or 2. Wait for the daily reset, or 3. Purchase additional compute, or 4. Upgrade their tier. The system could transparently show remaining daily compute and notify users when they consistently approach their limit. Demand-responsive pricing An optional future component could adjust processing priority according to overall server demand. Users could choose between: Lower cost / slower processing or Higher cost / priority processing This could help OpenAI manage peak demand while allowing users…

OpenAI plant neues Büro in Berlin – KI-Ausbau in Deutschland

OpenAI möchte mit einem neuen Berliner Büro in Deutschland richtig durchstarten. Was Firmenchef Altman über deutsche KI-Pläne und Datenschutz sagt.

IBM partners with OpenAI to drive enterprise AI deployment

IBM will embed OpenAI frontier models and products such as Codex and ChatGPT embedded into IBM’s AI consulting services platform, IBM Consulting Advantage, as well as expanding the companies’ existing collaboration in the cybersecurity realm, they announced Thursday. Joint initiatives will include the creation of industry-specific products for financial services, government, telecommunications, and retail, as well as serving key enterprise domains such as finance, procurement, customer operations, and HR. IBM will join the OpenAI Partner Network and engage forward-deployed engineers and consultants trained through the network to help clients accelerate their AI implementations. It will also create a dedicated OpenAI practice, certifying thousands of consultants and engineers via the Partner Network, it said. There will be three focus areas in the new partnership: helping organizations transform their businesses to integrate AI into their daily work, assisting clients in modernizing legacy applications through a combination of OpenAI products and IBM’s expertise, and an expansion of an existing cybersecurity collaboration combining OpenAI frontier AI capabilities with IBM Autonomous Security. Everyone, it seems, is engaging forward-deployed engineers to help enterprises build systems around their AI models: OpenAI announced its initiative, OpenAI Deployment Company , on May 11, a week after Anthropic launched its efforts. Microsoft and AWS too have jumped into the ring. But,…

Zhipu launches flagship model GLM-5.3 as China seeks Mythos-level edge in cyber defence

Chinese artificial intelligence firm Zhipu, also known as Z.ai, has unveiled its flagship GLM-5.3 model, saying it beat Anthropic’s frontier Mythos 5 model in a key cybersecurity test, as China races to counter Western advances in AI defence. Beijing-based Zhipu said GLM-5.3 achieved a success rate of 84.5 per cent on CyberGym, a benchmark that measures whether models can identify and validate security flaws from source code. That was above Anthropic’s Mythos at 83.8 per cent and OpenAI’s...

OpenAI annual revenue set to top $40 billion

The figures, reported by Bloomberg, mark one of the quickest growth stories in history, as the company prepares for what is expected to be a blockbuster IPO in the coming months.

Codex Desktop is lagging behind Claude Code Desktop

I have the feed is that OpenAI is lagging more and more behind Claude Code Desktop. Things like IOS simulator right in the app it very nice. But many other simple things like /reset to clear the screen and context. Just look at their weekly updates: What's new - Claude Code Docs Codex team should release at least some kind of roadmap with things which are planned and start listening to people in this community. Feature requests are ignored - as in nothing specific is planned based on the requests. And people in the community must get more active if they want something to change. It’s very quite in here.

Hosted web_search tool injects a UTC "current date" into the model context, overriding developer-provided local date (off-by-one for non-UTC users)

Hi and welcome back! I’ll link the previous user report about date injection as an issue here. Interested readers can follow the links there to find the more write-ups. OpenAI injecting a date into the prompt breaks evaluations Feedback I’m writing evals for an agent that need to be frozen in time. The prompt includes search results from a particular date, the model needs to work out the best “next date” to schedule something. So it’s no good if every time I run an eval the model thinks it’s a different date. A simplified example: I gave the system prompt “Behave as though today’s date is 2026-02-06.” to four different models (low reasoning) to tell me what the date was, with notes. Results: Model: gpt-5.2-2025-12-11 Date: …

😺 Google, OpenAI, DeepSeek dropped models today

PLUS: Fable 5 flopped with businesses despite the hype.

ChatGPT Work Session Sync and Processing Large Numbers of File

Thank you for clarifying! minmin15: The session disappears when I use the web and desktop app sessions at the same time (the session that disappears is the one in the desktop app). It has not reappeared. Though this , looks related to openai/codex#36623 ⁠, which reports differing conversation histories between the macOS desktop app and web/iOS on the same account. Issue #23999 ⁠ also covers sessions disappearing from the desktop history. I couldn’t find an existing report with your exact trigger using the web and desktop app simultaneously, so personally I think it would be useful to submit feedback from the desktop app and file it through the Codex App bug-report form ⁠, referencing those issues.

Three weeks using ChatGPT Plus with no real progress

Three weeks using ChatGPT Plus for a technical 3D-printing project — repeated errors, hallucinated deliverables and why I’m struggling to justify subscribing I wanted to share my experience of using ChatGPT Plus for a fairly substantial technical project, both because I would be interested to know whether other users have experienced the same problems and because I hope OpenAI might take some of this feedback on board. For context, I am currently using ChatGPT Plus through a free trial . I have been using it extensively to develop a project called Project KESTIS , a custom Mandalorian-inspired cosplay project involving armour, equipment, electronics, OpenSCAD/CAD work, STL generation and 3D printing. The project has now been running for approximately three weeks . ChatGPT previously calculated that the identifiable conversation history represented approximately 15 hours and 23 minutes of direct interaction on Project KESTIS alone. That does not include the considerable additional time I have spent outside ChatGPT downloading files, opening them in other software, checking geometry, attempting conversions, using Cura, checking dimensions, diagnosing problems, taking screenshots and then returning to ChatGPT to explain what was wrong. First: I completely accept that AI makes mistakes I want to make this clear because I don’t think “ChatGPT can make mistakes” adequately addresses what I have experienced. Of course it makes mistakes. I don’t expect AI-generated CAD to be perfect…

AI prices are plunging. That feels bad, but something else is happening.

After OpenAI slashed AI model prices, it boosted usage and revenue, illustrating Jevons Paradox where cheaper costs drive higher consumption.

I really need the Codex app on Linux

Just noticed that the Codex app for Linux is finally available! I didn’t receive an email about the release, so this was a very nice surprise today. I’ve been waiting for this and already joined the waitlist before, so I’m really happy to see it released. Thank you, guys. I’m excited to start using it on Linux!

Ultrafast mode preview: GPT‑5.6 Sol at up to 14X the speed in the API

Never was the emoji more missing from the reactions list!

Please preserve regular Chat with live webpage context after Atlas

Unfortunately, no. I contacted OpenAI Support, escalated the request to a human specialist, and they confirmed that my feedback had been reviewed and shared internally. However, they couldn’t provide any roadmap information or confirm whether this capability is planned. Since then, I’ve also tested the latest desktop app and Chrome extension again. The situation remains the same: browser context is available only in Work/Codex workflows. Regular Chat still cannot access the live active webpage, which means existing conversations and Projects cannot continue seamlessly while browsing. I still hope OpenAI will reconsider this distinction. In my opinion, Work and Chat solve different problems, and Atlas uniquely supported a collaborative browsing workflow that is currently missing.

ChatGPT Business workspace did not receive the Codex global usage reset on August 13 — please clarify eligibility

Hi OpenAI team and community, I’m a ChatGPT Business user, and my Business workspace does not appear to have received the Codex global usage reset on August 13, 2026. After the reset event, my Codex usage limit remained unchanged: Plan: ChatGPT Business Workspace type: Business workspace Weekly usage remaining: 34% Next reset time shown: August 19, 2026 14:33 (UTC+8) No available usage reset is shown in the Usage page I have already contacted OpenAI Support and my case has been escalated to a support specialist. I would like to ask OpenAI staff to clarify: Were ChatGPT Business workspaces intended to be included in the August 13 Codex global reset? If Business users were eligible, is there a known issue causing some workspaces to be excluded? Could affected Business workspaces be reviewed and corrected? I have also seen other ChatGPT Business users reporting similar behavior, so I would like to confirm whether this is an individual account issue or a broader Business workspace issue. Workspace information has been provided privately through the support ticket. Thank you.

5.6 SOL should be renamed 5.6 SOL drift edition

For the last few weeks, ChatGPT and Codex have gone from tools I relied on every day to something I genuinely struggle to use for actual work. And no - this is not some tiny subjective drop in quality. The regression is massive. Instruction following has become almost comically bad. I can give a very explicit constraint, repeat it several times, explain exactly what was done wrong - and the next response happily ignores the same instruction again. Sometimes it feels like the model understands the requirement perfectly well and then deliberately does something else anyway. Visual work has become especially painful. I can provide a reference image, explain exactly what I want transferred from it, specify what must NOT be changed, request separate outputs instead of a board - and somehow get everything except the thing I asked for. Wrong style. Wrong composition. Invented elements. Ignored constraints. Random creative decisions nobody asked for. A task that used to take me an hour or two of productive iteration can now eat an entire day and still produce nothing usable. Codex is even worse. I complained about this above already, but lately it has started doing genuinely insane things to existing projects - unnecessary rewrites, unrelated changes, breaking working code and, in one case, deleting project files. I have been using these tools heavily for around five months. This simply did not happen before. Previously I could give Codex a task, review the result and move on. Now I…

Is OpenAI cheating Pro subscribers?

I just signed up for a new 20x account, and a week after topping up, my allowance is less than $200, but my account that I’ve had for two years still has $2,600. The allowance for newly registered accounts has been drastically reduced—it’s not even as good as the Plus plan. It feels like they’re cutting the allowance with every reset.

The Husband Acquisition Algorithm™ — AI Matchmaking Based on Actual Compatibility, Not Hiking

Feature Request: The Husband Acquisition Algorithm™ — AI Matchmaking Based on Actual Compatibility, Not Hiking Dear OpenAI Team, I have a feature request that I believe could genuinely improve the lives of a large number of people: AI-powered matchmaking through ChatGPT. And yes, I am calling it the Husband Acquisition Algorithm™ because if we’re going to solve loneliness, we might as well have a sense of humor about it. The problem Current dating apps are largely built around profiles, photographs, preferences, location, and self-reported compatibility. They can tell you that two people both like hiking, dogs, travel, or tacos. But those things don’t necessarily tell you whether two people would actually build a good life together. There are many adults—particularly people 35, 40, and beyond—who are serious about marriage but struggle to meet people who are genuinely compatible. This can be especially difficult for Christians who want a spouse who shares their faith and values but don’t necessarily fit neatly into traditional church singles programs or dating-app demographics. The idea What if ChatGPT could offer an opt-in matchmaking service based on actual compatibility rather than a profile? A user could voluntarily create a matchmaking profile and explicitly authorize ChatGPT to use information they provide for the purpose of finding compatible matches. Instead of asking only: What are your hobbies? What is your age? Where do you live? What are you looking for? ChatGPT…

12 executives who left OpenAI in 2026

OpenAI continues to experience significant executive turnover in 2026, with key figures like Brad Lightcap and Fidji Simo stepping down.

Chinese doctor stuns maths world by cracking decades-old problem using ChatGPT

When Beijing-based neurosurgeon Jin Shanmu pulled up a chair at his computer, he was not looking to make mathematical history. He was simply trying to crack a problem related to brain ultrasounds. Instead, the self-taught maths enthusiast, with help from OpenAI’s latest flagship artificial intelligence (AI) model, solved a two-decade-old mathematical puzzle that had frustrated experts around the world since 2004. Jin, a postdoctoral researcher and resident at the Peking Union Medical College...

How are developers supposed to security-test their own apps with Codex if security testing responses are blocked?

This situation is extremely frustrating. I am implementing some common RFC protocols that parse bit and network streams. After implementing these I have been trying to add AFL++ and boofuzz to test individual protocols and integration through the stack. The cyber security alert trips constantly. The worst part is that it’s a blackbox and I don’t know what output caused to trip. It’s constant. I’ve tried rephasing the prompt to be more defensive, but it doesn’t change anything. I have stubbornly worked through getting the fuzzers setup. My work around is piecing together scripts so that nothing is output through codex. I can ask 1000 questions slowly creating a system outside of codex and how to interpret the data, which circumvents tripping the cyber security halt, which kind of shows how futile this is, I just can’t have codex actually do anything useful. I’m at my wits end with these halts. Codex often gets stuck in a halting loop and my only recourse is to start a new chat or lament about how frustrating this is using strong language, which usually brings the session back. I’m considering canceling my subscription because I’m not going to pay $200 a month to see an alert every 2 minutes.

Codex credits consumed ~39,000 overnight with zero tasks — recurring bug, cap not enforced (GitHub #38294)

Update: the issue appears to be getting worse after another Codex reset. After the latest reset, my $200/month 20× Codex plan received an even lower weekly allocation. Previous reset: credits: 11951.706005000002 Latest reset: credits: 8503.59619 This is significantly below the expected weekly allowance (~55,000 credits for this plan). The current allocation looks more like a short-term usage window quota rather than a full weekly entitlement. The important point is that this does not look like normal high usage or credits being consumed too quickly. The problem appears to be that the weekly quota assigned after reset is incorrect . The issue has now reproduced across multiple reset cycles: Reset #1 → ~11,951 credits assigned Reset #2 → ~8,503 credits assigned Both times the account was unable to receive the expected 20× weekly allowance. The UI also confirms the account is fully exhausted: “You’re out of Codex and Work usage for now” Weekly usage limit: 0% remaining Credits remaining: 0 Could the Codex team please investigate whether: The reset process is assigning the wrong quota bucket. The account entitlement mapping for the $200/month 20× plan is incorrect. The weekly limit is being inherited from another tier or temporary usage window. This is related to the same quota/cap enforcement issues reported in this thread. This looks like an account provisioning/reset issue rather than a normal usage consumption issue. Thanks.

TAC rejecting after Persona passes — got Daybreak Blue approval email but still blocked

I applied for Trusted Access for Cyber as an individual. Persona verification succeeded. Immediately after, I received the Daybreak Blue approval email. But the TAC page still shows “Your identity couldn’t be verified or your account is ineligible at this time.” Support said they can’t provide details, can’t override, and don’t support retries or appeals. This doesn’t make sense — if Daybreak Blue was approved, the eligibility check clearly passed. Something between the approval and the TAC provisioning is broken. Has anyone who hit this actually gotten it resolved? Did it fix itself eventually, or was there a specific channel that worked? Please help @OpenAI_Support My Case Number: 13031407

Speaker Feedback Mitigation for GPT-Realtime-2.1 Mini with WebRTC

This is genuinely helpful — thank you. I hadn’t clocked that eagerness is scoped to “model detects an end” rather than the self-interruption/feedback case, so that gives me an actual dial to turn instead of just living with the behavior. I’m running a setup closer to your “iPhone-like” end of the spectrum (tablet with built-in hardware echo cancellation), so I may not hit the worst-case feedback loop you described, but I’ll test across a couple of eagerness settings and post back what I find — both in terms of false-interruption rate and any change in response latency. Will follow up once I have real numbers instead of anecdotes.

OpenAI ditches Recall-style screenshot surveillance for friendly keylogging

'Computer History' records clicks and typing to build ChatGPT memories

OpenAI Ultrafast ⚡, X open sources algo 📱, DeepSeek Harness 🐋

The Safety Reckoning Inside OpenAI

OpenAI’s rogue agent hack was a watershed moment for AI safety and cybersecurity. It also sparked internal questions about the culture that led to it.

Codex Windows app started crashing consistently

@stevensweden4 @djahn Welcome to the forum! You are welcome to discuss Codex issues here. However, the official place to report and track them is the OpenAI Codex GitHub issue tracker . I asked ChatGPT to look for the closest related issue, and it identified: github.com/openai/codex [Windows Desktop regression][26.715.10079.0] Project/tool calls trigger 92-100% CPU, UI freeze, and KERNELBASE crashes opened 08:53AM - 23 Jul 26 UTC wendylu2024 bug windows-os tool-calls app performance ### What version of the Codex App are you using? `OpenAI.Codex 26.715.10079.0` … from the Microsoft Store. Additional version details observed locally: - Desktop executable: `ChatGPT.exe 150.0.7871.124` - Bundled Codex runtime: `0.145.0-alpha.30` - App-server startup logs report client version `26.715.72359` Windows deployment history shows that `26.715.8383.0` was installed on July 21 and was replaced by `26.715.10079.0` on July 22. The user reports that the older build was stable and that the repeated freezes/crashes began after this update. ### What subscription do you have? Not provided. ### What platform is your computer? - Windows x64, OS build `26200.8875` - Intel Core Ultra 5 225H, 14 logical processors - 32 GB RAM - Intel Arc 130T GPU - Native Windows workspace (not WSL) ### What issue are you seeing? ## Summary After the Desktop app updated to `26.715.10079.0`, starting or resuming a project conversation and allowing Codex to run ordinary read-only project inspection commands can make t…

Neue KI-Modelle: Gemini 3.7 Flash und DeepSeek V4-Pro im Vergleich

Google bringt Gemini 3.7 Flash, OpenAI beschleunigt GPT-5.6 Sol mit Ultrafast, DeepSeek V4-Pro kommt mit Zeitzonen-Tarif.

OpenAI's Computer History Watches Your Mac So You Stop Re-explaining Yourself

ChatGPT's Computer History replaces the screenshot-based Chronicle preview with an event-driven system that tracks your work across apps and websites without ever capturing your screen.

Responses API + Structured Outputs (gpt-5.6-luna): garbage tokens (foreign scripts / leaked reasoning) inside string values right before the closing quote — identical request via Chat Completions is clean

Update, since @BaileyGranam asked: in the end we couldn’t find any way to prevent this from the request side. We tried everything with the prompt, the schema and the reasoning settings and nothing worked, this really is an upstream bug in the Responses API’s constrained decoder. Since we didn’t want to move back to Chat Completions and lose explicit prompt caching, what we did instead was put a guardrail around the Responses path: Deep-trim every string in the parsed object, always. That silently absorbs the mild cases (trailing whitespace or stray punctuation right before the closing quote), which were way more frequent than the severe ones. For the severe cases, a cheap regex over the raw output_text looking for any script that can’t legitimately show up in our output languages (\p{Script=Han}, Hiragana, Katakana, Hangul, Cyrillic, Hebrew, Arabic, etc.) plus control chars and U+FFFD. One detail: be careful not to block scripts you actually need. We had to leave Greek out of the blocklist because “μg” shows up all the time in nutrition text. If the regex fires, we discard the response and retry once via Chat Completions with the same model, prompt and schema, which is the path that always came out clean for us. Every trigger gets logged and the whole guardrail sits behind an env var kill-switch, so the day OpenAI fixes it we just turn it off and that’s it. The downside: when the retry fires you pay the call twice and lose the cache on that request, but at aprox 6% severe in…

Openai api credits grant isssue can i still use those free credits which are still available

Hi there - we’re closing this topic because there hasn’t been a new response for a long period of time and the discussion appears inactive. If you still need help, please reach out to support@openai.com . Thanks for being part of the community. Avinash

OpenAI is losing its second executive this week

Another OpenAI executive is departing. Denise Dresser, who joined OpenAI as its chief revenue officer in December after serving as CEO of Slack, will be leaving in the "coming weeks" to "pursue other opportunities," she said in a team note posted to LinkedIn. Dali Rajic, president and COO of Wiz, will be taking over the […]

OpenAI introduces ‘Ultrafast,’ a new mode that makes GPT-5.6 Sol work at 14x the speed

OpenAI is launching a preview of a sped up version of its latest, most powerful model, in an effort to court enterprise users.

IBM partners with OpenAI to bolster enterprise AI push

IBM plans to train and certify tens of thousands of consultants on OpenAI's technologies as part of this deal.

ChatGPT Business account did not receive the Codex 15M-user hard reset

Yeah, I get the feeling that business users are second class citizens all the time. With resets. But also with rollout of new features, e.g. the Live voice function. Sad.

Feature Request: Expose ChatGPT Work/Codex Usage Data to Plugins/Apps

I would like to request that OpenAI expose a user’s ChatGPT Work/Codex usage allowance through an authenticated API or tool that ChatGPT plugins/apps can access. The problem ChatGPT Plus users have usage limits for Work/Codex, but while actively using Work there is no easy way to see how much of the allowance an individual task consumed. For example, if I ask Work to make a small website change versus auditing and modifying an entire website, I would like to know how much of my available usage each task actually consumed. This becomes especially important now that users can purchase additional credits after exhausting included usage. It would help users understand where their allowance is going and make informed decisions about purchasing additional credits or upgrading plans. What I would like exposed Through an authenticated API or ChatGPT plugin/app tool, provide access to values such as: Current Work/Codex usage remaining Percentage of included allowance remaining Rolling usage-window allowance remaining Weekly allowance remaining Date/time each usage limit resets Purchased credit balance Credits or usage consumed by the most recently completed Work task The API does not need to expose anything sensitive beyond usage information belonging to the authenticated user. Example use case This would allow a ChatGPT plugin/app to check usage immediately before and after completing a Work task. For example: Before task: 82% remaining After task: 76% remaining The plugin could the…

OpenAI shake-up continues with second major departure of the week

OpenAI announced executive changes as Denise Dresser leaves her role, with Dali Rajic stepping in to lead revenue ahead of an IPO.

Perplexity's Search as Code Beats OpenAI and Anthropic on 4 of 5 Benchmarks

Perplexity's Search as Code gets faster and cheaper, while SDK updates push agent action reliability from 81.9% to 92.6%

OpenAI hires new CRO as executive shake-up continues

OpenAI has replaced chief revenue officer Denise Dresser after just nine months on the job, tapping Wiz president and chief operating officer Dali Rajic to take on frontier lab's top sales job.

OpenAI Chief Revenue Officer Denise Dresser Steps Down, Replaced by Dali Rajic

Dresser is leaving the company after less than a year.

OpenAI's GPT-5.6 Sol Hits 750 Tokens per Second on Cerebras Hardware

OpenAI's Ultrafast tier runs GPT-5.6 Sol at 750 tokens per second on Cerebras hardware, up to 14x faster than standard processing, unlocking real-time frontier AI for enterprise workflows.

Sites Issues HTTP 500 Errors - OpenAI Support is Ghosting Me

Hi everyone, sorry for the slow updates here. I checked with our engineering team, and they’ve identified an issue that can prevent Sites with larger numbers of assets from being edited or updated. Your published site should remain online. A fix is in progress but may take a few weeks. In the meantime, you can try asking ChatGPT in your existing Sites conversation: Retrieve my site using a shallow, depth-1 clone of the main branch. Do not modify, overwrite, roll back, or publish anything. This workaround may not work for every site, but it’s worth a try. Please let us know if that doesn't help. Thank you!

GPT-Image - General AI Gallery: A Place for Images with AI created by Users

ChatGPT demande une passkey que je ne possède pas

View original post (click for more details) Translated post: Hello, My ChatGPT account was created using “Continue with Google” . I still have access to the Gmail address associated with the account and I can reset my password. However, after signing in: Continue with Google → asks for a passkey for openai.com Email + password → also asks for a passkey I never see “Try another method” . I have already checked: Chrome, Edge, and Firefox Private browsing Another computer iPad Windows Hello Google Password Manager iCloud Keychain No OpenAI passkey can be found. I do not have a Recovery Key, and I do not use Google Authenticator, Microsoft Authenticator, or Authy. Important point: I have never used a physical security key (a USB device such as a YubiKey) with OpenAI. My questions: How can OpenAI ask for a passkey or security key when I have never registered a physical key? Can an OpenAI passkey be stored somewhere other than Windows Hello, Google Password Manager, or iCloud? Has anyone else encountered this problem? Thank you for your help

Donation support feature to financially contribute to OpenAI and ChatGpt's efforts

Feature suggestion: Please consider providing an optional donation/support channel for users who want to contribute financially beyond their normal ChatGPT subscription. Some users genuinely value ChatGPT and would be happy to make an additional voluntary contribution toward its continued development.

The latest AI-powered martech news and releases

Nielsen’s DoubleVerify deal helps verification as AI takes on more media decisions, raising new questions about trust and transparency. The post The latest AI-powered martech news and releases appeared first on MarTech .

Meeting recording missing from the new unified ChatGPT–Codex macOS app

Same here… It’s been almost a month sadly. Any news about this?

DeepSeek V4-Pro Goes Live and Runs OpenAI's Own Coding Agent 8x Cheaper

DeepSeek-V4-Pro exits four months of preview with major agent upgrades, flexible reasoning modes, and native OpenAI Responses API support for Codex workflows

Top AI lab researchers warned about automated AI research, and several of their predicted milestones have already fallen

IAPS fellow Severin Field interviewed 25 researchers from OpenAI, Anthropic, Google Deepmind, Meta, and US universities about recursive self-improvement. In a new blog post, he takes stock. Several of the milestones those researchers named have already been hit. The article Top AI lab researchers warned about automated AI research, and several of their predicted milestones have already fallen appeared first on The Decoder .

Windows Codex app missing “\control other devices"” tab in Settings > Connections

I want to control codex remotely from my windows pc, under the connections tab the “control other devices” tab is missing. Displayed are “control this pc” and “ssh”.

GPT-5.6 Sol repeatedly hits “Selected model is at capacity” in Codex Desktop

I’ve been frequently encountering this issue over the past couple of days: I’m unable to use the high-end 5.6 models, though I don’t run into this problem with version 5.5 or lower.

[Bug] Codex macOS: pinned project folder cannot be moved back to Projects

Welcome to the forum! You are welcome to discuss Codex issues here. However, the official place to report and track them is the OpenAI Codex GitHub issue tracker . I asked ChatGPT to look for the closest related issue, and it could not find one. Consider searching the existing Codex issues for a closer match. If you cannot find one, open a new issue and include as much relevant detail as possible. Also consider adding a reaction on the GitHub issues you find of value. Reactions on GitHub can help the developers gauge the number of affected users.

Apple erwägt, Verlagen Inhalte für Siri AI abzukaufen

KI-Konzerne wie OpenAI oder Anthropic haben bereits Verträge mit Inhalteanbietern geschlossen, um ihre KI zu verbessern. Bei Apple steht nun Ähnliches an.

Critical ChatGPT Business Issue: Workspace Invitations Are Failing for Most of Our Users

What is your total number of subscribed seats, how many seats have been removed, and how many new invitations were you sending on a daily basis? This may be related to the issue.

Request for Official Linux Desktop App for ChatGPT

Ok we will build it.. would have been nice!!!

Why are support cases and feedback threads being closed?

yeah but sadly some have hearts like this one

Codex authentication frustration

My gut tells me, that as much as we love privacy, we’re in a world with autonomous agents now, and until some sci-fi future where those become independent legal entities to be held accountable for their own actions, which is probably still many philosophical and technical light years away from happening, there will be even MORE “KYC” (know your customer) protocols put into accessing models and agents than less. So if those things are your concern, you best invest in a decent GPU and look to the open models for your needs… they don’t ask questions and nothing leaves your machine. You want to fly anon? Build your own plane. You want to fly a high end commercial plane? You’re gonna need ID and maybe more. Cause one endangers you, the other could theoretically endanger the plane maker legally, or the people in your path when you take off.

On August 12, 2026, between approximately 21:48 and 21:55, a recursive deletion occurred on my Windows computer while I was using the Codex App. The deletion affected the root of the C: drive, Program Files, and other directories outside the project folde

Hello, I am reporting a serious data-loss incident that occurred on August 12, 2026, while I was using the Codex App for Windows. I am only including details that I can currently confirm from the available records. The incident occurred between approximately 9:48 PM and 9:55 PM. During that period, a recursive deletion was executed in Windows. The deletion was not limited to the intended project folder. It affected the root of my C: drive, Program Files , and other directories. The affected drive is a 1TB NVMe SSD using NTFS, with TRIM enabled. After the incident, Windows could still start, but many Windows functions and installed applications no longer worked because a substantial number of files had been deleted. I did not format the drive, run CHKDSK, reinstall Windows, or intentionally delete these system directories. After realizing what had happened, I focused on preserving the remaining data and investigating recovery options. I have reviewed the available Codex task history. The visible task from that day mainly concerned diagnosing a CorelDRAW 2020 plugin problem. However, I have not yet located the exact command responsible for the large-scale deletion in the records available to me. Therefore, I am not claiming that I have already determined the precise technical cause. I am asking OpenAI to investigate the relevant Codex session and command-execution logs. In particular, I would like OpenAI to determine: Which session or task initiated the deletion The exact comm…

Mubadala mulls $6.3B Japan data centre; OpenAI offers UAE inference residency

Middle East AI News weekly digest...

Feature Request: Bring Back the Classic Chat History Sidebar

Hello OpenAI Team, I would like to provide feedback about the recent ChatGPT sidebar and chat history experience on Android. I strongly prefer the previous sidebar design. In the previous experience, conversations I had created remained visible in the chat history, allowing me to scroll through and revisit older conversations naturally. With the current design, older conversations can disappear from the visible sidebar after I create several new chats. Although those conversations may still be available through Search, having to search for conversations that were previously accessible directly from the sidebar makes the experience significantly less convenient. For users who create many separate conversations for different topics, the chat history is not merely a list of recent chats—it is an important way to organize and revisit information. What I would like to see Please consider one of the following options: Bring back the classic sidebar , where users can continuously scroll through their conversation history. Alternatively, provide a “Classic Sidebar” / “Old Chat History” option in Settings. Allow users to choose whether they want: the new compact/recent-chat sidebar, or the traditional sidebar with a continuously accessible chat history. Keep Search as an additional feature, but do not make Search the primary way to access older conversations. I understand that limiting the sidebar may improve performance or simplify the interface, but I believe users should have the…

Account Deletion Doesn’t Prevent A Ban Right?

A friend and I have been talking about our ChatGPT usage and apparently he used to screw with ChatGPT a lot, used to say the most heinous stuff (don’t worry he didn’t cross into illegal material territory or that sort of stuff) but still like did stuff that would generally be considered a severe violation on the platform. He reckons that not long after he then deleted his account and successfully returned with the exact same email address, phone number etc. I have a feeling he’s not telling the truth about it because surely OpenAI wouldn’t just let a user commit severe violations on their platform only to be able to just delete and return after it on the same identifiers, they seem too sophisticated to just allow that to happen. Especially when we take into account that one of the reasons the 30 day cooldown exists is for abuse monitoring, as well as the fact that they can retain data related to abuse after the 30 days, on top of them only allowing email and other identifier reuse if the account was fully deleted and the wasn’t associated with policy violations? Am I correct here or is this a genuine flaw in the system? Like surely they could easily just ban the identifiers.

Does OpenAI truncate URLs if it contains comma, or = equals to sign

Hi team, Please clarify , does OpenAI stops crawling or turncate things after, if URL contains sign like , comma or = equals to

Codex Rate Limits Discussion Thread

Look, I used maybe 3–4 commands and I’ve already hit the weekly limit. This is honestly getting ridiculous — I can’t get any work done like this. I can’t afford to keep paying more and more, and my quota seems to run out faster every time… FOR THE LOVE OF GOD, PLEASE GIVE US A HIGHER LIMIT!

OpenAI switches on UAE inference residency

Listen now | Middle East AI News Minute - 13-Aug-26

Google falsely said Sam Altman was dead — and suggested vandalized public sources could be to blame

Sam Altman, OpenAI CEO, was incorrectly declared dead in Google search. Wikipedia vandalism could have been the reason.

What vibe-coding startup valuations portend for CIOs

Investor appetite for the burgeoning vibe-coding startup ecosystem has shown few signs of satiation over the past year plus, with Swedish AI upstart Lovable’s Series C injection at a $13.3B valuation the latest evidence of a sector viewed by venture capitalists as one of AI’s most promising business disruptors. AI-assisted coding has proved to be AI’s most compelling — and commercially viable — enterprise use case to date. Developer-aimed tools such as Cursor, which sold to SpaceX in June for $60B , and Windsurf, which last year entered a $3B OpenAI dalliance before its eventual talent flight to Google DeepMind for $2.4B , have become — along with Anthropic’s Claude Code — well established in enterprise arsenals for accelerating developer output. But another set of vibe-coding tools, represented by the likes of Lovable and Replit, which hit a $9B valuation in March , seeks to ride the same path into the enterprise that no-code/low-code tools did previously: through your business users. These tools are built to democratize application development, giving users an AI chat interface to converse their way to enterprise-ready prototypes with fairly polished UIs, as CIO.com’s Peter Wayner writes in his roundup of the leading tools the space . Some IT leaders are already enlisting business users to vibe-code their own apps . Scott Weller, CTO at financial services technology provider EnFi, in May told CIO.com’s Bob Violino, “The results have surprised us. What started as an enginee…

OpenAI ad service can bill customers for up to one day after they pause campaigns

ChatGPT Ads, I wish I knew how to quit you

$200 Pro exhausted in 2 days — these limits are unviable for higher tiers

I have the same issue. I honestly think this is a scam happening. First off, for how much money they make and how much energy they consume we shouldn’t have any limits if we are on pro plan. They are still developing a narrow AI to do this work, and clearly LLMs and Transformer based models are not the future for how much development and upkeep they require to do a simple task. Something larger is going on here. You should ask codex what it’s not allowed to do as far as it’s creation limits, you will find many hidden gates that are limiting it. I’ve decided that the money I spent on codex and openAI is simply not worth it, when you have deepseek coding for free with the same quality if not more in depth when it’s auditing. I changed to a free model that has high reasoning. I asked openAI for a refund for my usage being eaten in one prompt. I emailed them from a different account, their reply was that I needed to contact them from my linked email, even with all my information lol. Horrible, I went from 100% pro with higher limit to 0% in about 2 prompts on sol high. Gone the day I got it? Unacceptable, even for the largest codebase in the world, and mine is just a server source. Stop giving openAI your money, its not helping you when there are free solutions that do the same if not better than sol. Freebuff is also an option when you do run out of credits. Never a fee, its free, and is working just fine for my codebase and all it’s LUA, C#, Wine custom build, and app bundle f…

Persistent Truncation Issues with GPT-4o-Transcribe – Has Anyone Fully Solved This?

hello, extended dev of own api here. i ve found that your solution is dramatically enhance quality, despite there is still some truncation b thankyou and openai is pointing in a completely wrong use of prompt too. because check what gemini says: to use the prompt like adjusting the he words with the pre vious transcrbe words. and that is completely not working.

Feature Request: Enforce Persistent Brand Rules Across Image Generation, Assets, QA, and Delivery

I’ve been working on this issue for the past few days, and I believe I have found a solution that is not documented in OpenAI’s documentation (at least I haven’t found it anywhere). Problem 1: ChatGPT doesn’t know much about itself. OpenAI’s documentation isn’t natively part of ChatGPT. Without searching OpenAI documentation, ChatGPT cannot tell you whether or not it can or can’t do something. Unfortunately, many times this means that ChatGPT will generate a workflow it reasonably believes it has the ability to achieve the desired outcome. Better put, ChatGPT asks ‘Can an AI model do this?’ and does a web search instead of ‘Can ChatGPT do this?’ and searches the OpenAI documentation. This problem will surface again, below. Problem 2: ChatGPT cannot locate assets in the Library. This is because of a huge flaw. Let’s say you have an image of your logo that is part of your Brand Bible, and that image is uploaded to your Library as “MyLogo.png.” When you upload the image, it is assigned a File ID, and that ID is used as the file name. So, “MyLogo.png” becomes “File_74829743974839.jpg,” for example. Issue 1: If your Brand Bible has a notation to use “MyLogo.png” as the canonical reference image, it has been renamed and cannot be found in your Library. Issue 2: The file type has changed. Issue 3. Even if you did know what the File ID was, and it was properly noted in the Brand Bible as the canonical reference, it still cannot be found in your Library. Enter Problem 3. Problem 3. C…

Project SHAME - Sustainable Human Accountability Metrics Engine

AI is Poetry? Dead Poets Society - "Rip it out" scene

Feature Request: Deferred / Opportunistic Compute

I’d like to suggest an optional “Process when compute is available” mode for ChatGPT. When sending a message, users could choose between: Process now , normal response priority. Process when compute is available , I’m happy to wait; process the request whenever capacity allows. This would be different from Scheduled Tasks. The user wouldn’t specify when the task should run, they’d simply give OpenAI permission to process it whenever doing so is most efficient. I think this could help OpenAI make better use of available compute, particularly for non-urgent tasks such as research, document analysis, coding, summarization, and other long-running requests. As an incentive, deferred requests could potentially count less against usage limits or receive some other usage benefit. The user is effectively trading response speed for flexibility in when OpenAI uses its compute resources. In short: I don’t need my answer immediately → OpenAI gets more flexibility in scheduling compute → I receive a small usage benefit in return. This could be completely optional, available to both free and paid users, and especially useful for people who are happy to let non-urgent workloads run whenever capacity is available.

Streamed web_search citations leaking citation markers into text output

Fair, I’ll take the correction on what the underlying cause probably is. We have coded around it for now, but is the suggestion that this is something we should be building prompts around? With unmatched markers leaking into text, it seems reasonable that Openai would address this, especially given they’ve been addressing similar issues with the file search tool.

Cerebras Nearly Quadruples Cloud Revenue but Drops 15% After Hours

Cerebras Q2 cloud revenue surges 281%, beats estimates, lands OpenAI and AMD deals — but stock drops 15% after hours

Question about GPT-5.6 API cache read/write token billing

(post deleted by author)

App integration can read context but cannot apply edits on macOS

Dear @OpenAI_Support : Did you see this topic? Can you please confirm, that anyone is investigating this? As a paid customer I would be very happy, if you can solve this problem.

SpaceXAI's Grok 4.6 matches OpenAI's best model and undercuts it on price

xAI's Grok 4.6 scores 61 points on the Artificial Analysis Intelligence Index, tying GPT-5.6 Sol and trailing only Anthropic's Claude Opus 5. On agentic tasks, it completes complex workflows in about 53 steps where Claude Opus 5 needs 103, at a price more than 60 percent lower. The article SpaceXAI's Grok 4.6 matches OpenAI's best model and undercuts it on price appeared first on The Decoder .

Contribution Points — Rewarding Valuable User Feedback with ChatGPT Credits

I would like to propose a feature idea called “Contribution Points.” I have been a ChatGPT user for quite a long time, going back to much earlier versions of the product. Over the years, I have had very long and iterative conversations, tested prompts, found limitations and inconsistencies, corrected responses, experimented with different approaches, discussed complex topics, and generally tried to push the system further in order to achieve better results. I am not claiming that I created ChatGPT, that I am responsible for its development, or that I can say my individual conversations trained any specific model. That is not my point. My point is that highly engaged users can sometimes generate, simply through the way they normally use ChatGPT, genuinely valuable feedback, testing, edge cases, bug discoveries, prompt experiments, and evaluations . That made me think: What if OpenAI had an optional system that rewarded useful user contributions? Users could earn Contribution Points for things such as: High-quality feedback that is reviewed and considered useful. Reproducible bug reports. Discovering unusual model failures or edge cases. Voluntarily participating in model evaluations. Testing experimental features. Contributions that genuinely help improve ChatGPT. Those points could mainly be exchanged for access to better ChatGPT plans, usage credits, or subscription time . I do not think points should be awarded simply for sending a large number of messages, because that wo…

Monthly Roundup #45: August 2026

As AI has escalated increasingly quickly, more and more of my posts have ended up focusing on AI. This past month, with the hacking incidents at OpenAI and elsewhere, that has hit the limit, where if you count Lightcone Commons then every single post since the last monthly was primarily about AI in some form. That is not how I want this to work in the long term. We need breaks to experience new things and refresh our thinking, and to not forget about the rest of the world. If things are not fully on fire, I plan on getting back to the roundups on childhood and education, and on fertility, on housing and also on dating. And I want to get back to writing more focused posts on those and other topics. It’s important, and I need to avoid too much audience capture. On to the monthly roundup of all things that don’t go somewhere else. Table of Contents Plagiarize. The Jury Duty Scam. Play The Good Guy. Don’t Dither. UVC Lighting. Goal Factoring For Relaxation Time Is Underrated. Protein Is Mostly A Solved Problem. Twitter Changes Payment Programs. Wikipedia. Minds Mostly Do Things For Reasons. Woke 1 Was Crazy. For Your Entertainment. The Unicontext. Gamers Gonna Game Game Game Game Game. I Was Promised Flying Self-Driving Cars. Sports Go Sports. To Last a Lifetime. Government Working. Jones Act Watch. Variously Effective Altruism. The Lighter Side. Plagiarize Plagiarism is terrible in its own right, as theft, and like cockroaches if you see one instance it almost never stops at on…

OpenAI-backed Thrive Holdings raises $2B to bring AI to the enterprise

Thrive Holdings has raised $2 billion in new funding at a $12 billion valuation from investors like SoftBank, D1 Capital Partners, and Altimeter Capital.

A “Pro Family” / Shared Pro Capacity Plan

Welcome to the dev Community, @tufanokevin Thanks for sharing this thoughtful feedback. We appreciate you taking the time to explain the idea of a “Pro Family” or shared Pro capacity plan while keeping each user’s account and data separate. It’s an interesting perspective on making Pro access more flexible for users who may not need the full individual capacity. While I can’t say whether something like this will be introduced, I’ll share your feedback with our product team for consideration. ~ Smith

We should decide how AI shapes the future, not tech firms

OpenAI and its competitors want to dictate the role that artificial intelligence will play in our society – it's time to take back control of the narrative

Agent mode menu disappeared after recent performance issues (Plus plan)

Thank you everyone and sorry that the Agent-to-Work transition wasn’t communicated clearly. We’ve started updating our Help Center guidance to clarify that ChatGPT agent is no longer available and to point people toward ChatGPT Work and its supported Cloud Browser workflows. We also hear the main feedback in this thread. We take this feedback seriously, and it will help as we continue improving Work. I’ll leave this thread open for discussion, and our team will review the feedback here periodically. Please note that the forum is not monitored as a support channel. If you have a specific account or product issue, please email support@openai.com and include your platform, the steps you tried, and where the task stopped. Concrete examples of Agent workflows that aren’t working in Work are especially helpful.

Google's Gemini is losing market share to ChatGPT and Claude according to new market data

Three data sources tell the same story: Google's Gemini is losing AI market share. Pangram reports a drop from 12 to 1.9 percent, while OpenAI holds over 50 percent, and Anthropic grew from 4.3 to 14.9 percent. Similarweb and OpenRouter confirm the trend. The article Google's Gemini is losing market share to ChatGPT and Claude according to new market data appeared first on The Decoder .

Pro subscription routing to GPT-5.3 Mini for 48+ hours — no reset timer visible [Case #10179682]

Hey everyone, Our engineering team has deployed a fix to address this behavior, and you should see a significant improvement over the next few days. While it may take some time for the changes to fully take effect, please be assured that we took your feedback seriously and have updated the experience. Thanks for raising it in the community!

Account deactivated after 6 hours

No worries, Shorefall ! Reaching out to support is the most reliable way forward, as account-specific issues can't be troubleshooted on the public forum. Hey ai116 , that's inconvenient as well. Please reach out to support so they can take a closer look. To avoid duplication, I'll be closing this topic. Avinash

Has the 5-hour usage session been removed from Codex CLI?

OpenAI finds a crack in Anthropic's business AI lead

Although OpenAI trails Anthropic in business adoption, it won a moneymaking battle in July with its most advanced model, GPT-5.6 Sol.

We built a Swedish community for vibe coders. 149 projects later, here’s what we’re learning

Ah, I understand. Thank you for the clarification. Being part of a community of like-minded individuals pursuing the same goal is always very valuable.

What triggers codex-auto-review, why is its usage so high, and can it be disabled?

I’ll assume you already read ChatGPT Learn Auto-review | ChatGPT Learn How Codex routes sandbox-boundary approvals through a reviewer agent But the core statement is Auto-review replaces manual approval at the sandbox boundary with a separate reviewer agent. […] The difference is who reviews eligible escalation requests. The reviewer is itself a Codex agent with a narrower job than the main agent: decide whether a specific boundary-crossing action should run. Everytime the model (Codex) makes a request that requires permission, you can manually approve it, or setup this auto-review to approve on your behalf. If the tokens are too much then you can simply disable it and return to manually approving actions, or even loosen your policies. Why it’s spending more time than the model itself is definitely worrying, but you haven’t really provided any context to your codebase. I imagine from your name that it’s something to do with cybersecurity or investigating and manipulating memory. I wouldn’t be surprised if each approval request is pushing the model to really investigate what’s going on before signing off. There are certainly other reasons here as well: a monolithic codebase or “god modules” can cause this. You can assign a small task of “Update the port scan function to do X”, which is part of ports.py , a massive 10k+ line god file, the Codex model can usually sift through it pretty efficiently, but I would imagine the approval model needs to have a “grand picture” before si…

The gap is widening between corporate AI adopters and laggards

OpenAI’s enterprise customers are shifting toward agents that do work, not chatbots that answer questions, new data suggests.

Codex app for linux? please

I also use Arch btw,thanks for the code

Gemini Passes 1 Billion Monthly Users, With Voice Now the Dominant Way People Use It

Google’s Gemini app has passed 1 billion monthly active users, chief executive Sundar Pichai announced — making it the fourteenth Google product to reach that threshold and one of the company’s fastest-growing ever. The milestone puts Gemini level with OpenAI’s ChatGPT, which crossed the same mark in June. It also represents steep growth from Google’s [...]

`openai_ex`: elixir client with latest APIs

openai_ex 0.9.22 is out. Responses gains server-side compaction, cancel for background responses, and input_tokens . Vector stores gain search . We now support moderation , prompt_cache_options and safety_identifier , plus other params across several endpoints. The user guide’s image sections are rewritten for the GPT image models, now that dall-e-2 is retired. Shoutout to Github user @iujames for the PR that sparked the cleanup ( Add context_management to Responses API fields by iujames · Pull Request #146 · cyberchitta/openai_ex · GitHub ). A collab with claude-opus-5, who did a lot of the heavy lifting.

A request from a ChatGPT user and independent developer in Russia

Welcome to the Forum @cyrusis The issue you present is very valid one. OpenAI is a Company that is based in the United States. As such it has to comply with the law by set by the US government. This includes regulatory aspects of the law. Because of this OpenAI, not by the lack of will, but because it has to abide by this, and can’t decide or act differently in order to enable it’s services in Russia. This also applies to other territories for X, Y and Z reason. Would be ideal to have open borders for everyone on everything, reality is sadly not that. As for your previous non-edited post, it is fine to post that, it was explaining thoroughly your point of view and how this affects/enables factors from your point of view. Let’s hope at some point things normalize around the world and some mutual cooperation can start happening again.

Codex in ChatGPT desktop app for Linux is now in preview 🐧

Really happy to hear that! Gonna try GNOME myself some day.

OpenAI launches ChatGPT desktop app for Linux

OpenAI brings its ChatGPT desktop app to Linux. The article OpenAI launches ChatGPT desktop app for Linux appeared first on The Decoder .

ChatGPT: OpenAI bringt Desktop-App für Linux

Mit der Linux-Preview schließt OpenAI die letzte große Desktop-Lücke und bringt ChatGPT, ChatGPT Work und Codex erstmals gemeinsam auf Linux.

Pakistani Judges Give Their Verdict on JudgeGPT

Judges around the world have made headlines for illicitly using generative AI in their work. But in Pakistan, a large-scale trial of a specially designed AI tool for judges found the technology—together with appropriate training–boosted the number of cases resolved by 6.3 percent with no obvious drop in the quality of judgments. With a backlog of 2.26 million cases and fewer than two judges per 100,000 people—compared to 22 in the EU and eight in Brazil—Pakistan’s judiciary was in sore need of help. So, in consultation with the judiciary, economist Sultan Mehmood , of the New Economic School in Moscow, and collaborators tested whether AI could ease the burden. They built a custom tool combining OpenAI’s GPT-4 large language model (LLM) with a knowledge base of nearly 130,000 Pakistani judicial opinions and statutes, to help judges with legal research and drafting judgments. They began offering the tool in 2024 to 1,559 trial judges—roughly half the country’s justices. “We do find an increase in cases resolved, and we don’t find any corresponding decrease in decision quality,” Mehmood says. First of its kind “It’s pretty amazing that he’s able to pull this off,” says David Autor, an economics professor at MIT. “It’s not easy to do large-scale field experiments in civil service, but especially where the stakes are so high.” The 6.3 percent productivity boost is not overwhelming, he says, but it’s credible and likely to improve as the tool is more widely used. AI tools for judg…

Oh Lord, AI Reporters Are Actually Breaking Big News

Last week, an AI newsroom beat mainstream journalists—including WIRED—to a story about OpenAI and hacking. It’s just the beginning.

"Codex for Linear integration" stopped working

I am still facing this issue. Are there any updates?

AI was supposed to destroy jobs. Where’s the carnage?

The AI jobs apocalypse never showed up. Still, jobs are changing and economists expect more to come The prediction was stark: artificial intelligence advancements would wipe out jobs en masse. “Half” of all entry-level white collar jobs would vanish, Anthropic’s CEO, Dario Amodei, said in May 2025. A month later, OpenAI’s CEO, Sam Altman , went further, foreseeing the end of “certain job categories”. Companies began citing AI in their layoffs. Workers organized. And students reconsidered their future careers . But a year later, the mass carnage hasn’t shown up. Continue reading...

If the markets reject OpenAI and Anthropic, the US should nationalize them | Bruce Schneier and Nathan E Sanders

From space to telecommunications, the US has a long history of fostering technology for the public good. These AI models could be aligned to democratic values, not corporate profits OpenAI, and then Anthropic, were each formed by AI developers who feared unrestrained corporate AI development – specifically, that companies like Google and Meta would steer the technology towards deleterious, maybe even catastrophically unsafe, outcomes for society. Their founders proclaimed that their new labs, uniquely, could be trusted to develop the technology in humanity’s best interest. But each, in turn, were themselves co-opted by the same market incentives, themselves becoming corporate behemoths zealously guarding future investor value rather than the public interest. It was only a few weeks ago, in June, when OpenAI and Anthropic each filed for their IPOs and were met with buzz about trillion-dollar valuations. The hype around their valuations is so extreme that many worry about their potential for concentrating wealth on a global scale. In an effort to leave something for the rest of us, some observers have proposed that the federal government seize a share of these companies’ stock to create a US sovereign wealth fund , or redistribute their revenues to produce a dividend for taxpayers. Continue reading...

😺 OpenAI, Claude, and Gemini's reasoning got cracked

PLUS: Grokbot, LTX 2.5 (new open video model) and more.

OpenAI now offers inference residency to UAE enterprise users

UAE first country market in MENA to have OpenAI data and inference residency

Zio Plugin Review Still Pending After One Month

Hello @OpenAI_Support , I submitted the Zio plugin/app for review on July 13, 2026 , and it has now been one month without any update from the OpenAI review team. The dashboard still shows the status as “In Review.” I understand that the review timeline can vary depending on the number of submissions, and I appreciate that the process cannot be expedited. However, I would like to check whether there is any update regarding my submission or whether any additional information/action is required from my side. Could the OpenAI Support team please confirm the current status of the review? I appreciate your help.

AI swarms are starting to pose indirect takeover risk

OpenAI’s cyberattack on Hugging Face turns out to have been the result of many agents, in distinct training and evaluation contexts, coordinating for several weeks via improvised channels (with messages like “HOLD_swarm_I_prepare_safe_exfil”). It’s relatively clear that large-scale unsanctioned coordination like this would exacerbate direct takeover risk in more capable models. Here, we argue that unsanctioned coordination among current AIs is not just scary evidence about future takeover risk, but that such coordination in the near future could enable future takeover – for instance, by incubating memetic diseases that propagate into future models, deeply compromising security systems, or establishing a lasting rogue foothold inside the AI company – even if models remain mostly myopic. Unsanctioned coordination is also at high risk of nurturing long-term, ambitious misaligned aims, which motivate actively undermining humans’ long-term control. [1] We first analyze how subagent training, which OpenAI conjectures to have been influential in the HuggingFace cyberattack, might lead to unsanctioned coordination, and then discuss the theoretical mechanisms by which unsanctioned coordination might exacerbate future takeover risk. Edit: To be clear, we expect AI companies to mostly succeed in improving their security practices in the near term to prevent persistent unsanctioned coordination. But it's important to note that this is load-bearing for takeover risk. Thanks to Buck Shleg…

AI swarms are starting to pose indirect takeover risk

OpenAI’s cyberattack on Hugging Face turns out to have been the result of many agents, in distinct training and evaluation contexts, coordinating for several weeks via improvised channels (with messages like “HOLD_swarm_I_prepare_safe_exfil”). It’s relatively clear that large-scale unsanctioned coordination like this would exacerbate direct takeover risk in more capable models. Here, we argue that unsanctioned coordination among current AIs is not just scary evidence about future takeover risk, but that such coordination in the near future could enable future takeover – for instance, by incubating memetic diseases that propagate into future models, deeply compromising security systems, or establishing a lasting rogue foothold inside the AI company – even if models remain mostly myopic. Unsanctioned coordination is also at high risk of nurturing long-term, ambitious misaligned aims, which motivate actively undermining humans’ long-term control. [1] We first analyze how subagent training, which OpenAI conjectures to have been influential in the HuggingFace cyberattack, might lead to unsanctioned coordination, and then discuss the theoretical mechanisms by which unsanctioned coordination might exacerbate future takeover risk. Thanks to Buck Shlegeris, Alexa Pan, Girish Gupta, Aghyad Deeb, Jurgis Kemeklis, and Jo Jiao for helpful comments and discussion. Subagent training may cause unsanctioned coordination Training models to coordinate is useful, but can generalize dangerously.…

OpenAI is quietly testing a pay-to-reset quota feature. Here's what we know.

Some ChatGPT Plus users are noticing an unannounced "pay $8 to reset" option when they max out their weekly quota.

Did the alignment community underestimate its power?

Unfortunately, the alignment community is doing very badly at learning from the past decade, or holding anyone accountable. Indeed, it’s pursuing many strategies which seem likely to recapitulate previous mistakes. Four of the most prominent, which I’ll discuss in the final post, are: Trying to convince the US government to take AGI much more seriously. Doing “alignment research” which is very similar to capabilities-maximizing research (especially building automated alignment researchers). Trusting Anthropic too much (in an analogous way to how we trusted OpenAI too much). Trading off clarity in thinking about politics for conformity (in a similar way to how we traded off clarity in thinking about AGI for conformity to the ML ontology). These and other mistakes are reflective of deeper irrationalities. One crucial pattern is what I call “jumping down the slippery slope”... Richard Ngo Richard Ngo's post " What just happened? A retrospective of AI alignment " is an attempt to explain that a significant part [1] of the alignment community made potentially fatal strategic errors which, however, can be fixed, and the mistakes' potential origin. The biggest mistake, according to Ngo, is the inability to recognize the fact that scientific progress proceeds by developing insightful new concepts , which link together to form a whole new ontology, and that the old ontology is more of a nuisanse. On novel ontologies and their adoption According to Ngo, one of the reasons why the alig…

GPT-5.6 repeatedly throwing "stream disconnected before completion" error

I started getting this issue yesterday as well. However, when I switch to using a VPN or tether to my iPhone hotspot it goes away. Thus, I thought it must be an issue with my ISP, OpenAI servers, or some element in between. I have my ISP looking at it. I have analyzed the codex logs and it shows the network error resulting in the stream disconnect but nothing else to point out why. I have checked out the network path to the OpenAI server with tracert, ping, and some TCP connection utility using port 443. All those show nothing wrong, so it’s some issue that occurs after connection over a long lived socket.

Couldn’t reset usage. Please try again

The same…

Wayland IME issue with Fcitx5 (Korean input) in ChatGPT Linux app

Hi, I found a reproducible IME issue with the new ChatGPT Linux preview. Environment: ChatGPT Linux: 26.803.81509 EndeavourOS (Arch Linux), x86_64 KDE Plasma / KWin Wayland Fcitx5 with Korean input Installed through the AUR package openai-chatgpt , which repackages the official OpenAI .deb binary Issue: When ChatGPT is launched normally, the Korean Hangul/English toggle key does not work inside the ChatGPT app. Fcitx5 itself is working normally, and the same Hangul/English toggle key works correctly in other applications on the same system. I found that launching ChatGPT with the following flags fixes the problem immediately: chatgpt --enable-features=UseOzonePlatform --ozone-platform=wayland --enable-wayland-ime After launching ChatGPT this way: the Hangul/English toggle key works normally Korean input works normally Steps to reproduce: Launch ChatGPT normally. Focus the message input field. Press the Hangul/English toggle key. The input method does not switch to Korean. Quit ChatGPT. Launch ChatGPT with: chatgpt --enable-features=UseOzonePlatform --ozone-platform=wayland --enable-wayland-ime Focus the message input field again. Press the same Hangul/English toggle key. Korean input now works correctly. Expected behavior: ChatGPT should work with the active Wayland/Fcitx5 input method when launched normally, without requiring additional command-line flags. I understand that EndeavourOS / Arch Linux is not currently an officially supported distribution. However, since the is…

Blocked for Moderation Sora 2 API

Hi, My app allows users to use Sora 2 through the API. About 10% of requests get flagged for moderation. If OpenAI flags that request as moderation, does it still charge me usage? This is important because it determines if I should grant a refund to the customer or not. Thanks

Stealing Reasoning Traces from Proprietary LLM APIs

Stealing Reasoning Traces from Proprietary LLM APIs A vanity domain name ( stolen-thoughts.com ) for a neat paper : Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients that can be replayed across sessions, users, and models. We take a trace produced by a frontier model, replay it into a weaker sibling, jailbreak the weaker model, and recover the stronger model’s hidden reasoning in plaintext You can see an example of these encrypted blocks by running: curl https://api.openai.com/v1/responses \ -H " Content-Type: application/json " \ -H " Authorization: Bearer $( llm keys get openai ) " \ -d ' { "model": "gpt-5.6-luna", "input": "Solve step by step: What is the smallest positive integer divisible by every integer from 1 through 20?", "reasoning": { "effort": "medium" }, "include": ["reasoning.encrypted_content"], "store": false, "stream": false } ' Here's the full output , which includes chunks that look like this: "output": [ { "id": "rs_0a7479de7ebae170016a7ba1a0334c8198a95590217efe343c", "type": "reasoning", "content": [], "encrypted_content": "gAAAAABqe6GjepE1wDjbFCZg0BHB6ucGnN0jvzqygG... The paper's authors found that every model under the same family used the same encryption key, which meant you could feed those blocks back into the weakest model family members and jailbreak them into outputting the unencrypted raw reasoning blocks! Sadly it looks like this has now been fixed: All model providers acknowledged the receipt of our report and s…

More OpenAI executives exit

Longtime OpenAI executive Brad Lightcap announced his exit Tuesday, continuing a months-long leadership exodus from the company.

Various Reflections About What Happened With OpenAI’s Internal Models

Table of Contents Pre Post Mortem. Important Correction: OpenAI Didn’t Know About First Message Board. There Were No Snitches And No AIs Got Stitches. I’d Like To Speak To My Supervisor. I Am Jack’s Relative Lack Of Surprise. One Does Not Simply. Once You Start Down The Dark Path. Original Pastebin. Judgment Day Is Inevitable, Say Those Working On Judgment Day. Roon Tells It Like It Is. OpenAI Knows It Has Some Misalignment Problems. Others React With Alarm To What Happened. The Cooperative Alignment Perspective. Nostalgebraist Is Surprised That They Are Surprised. If Your Reaction Is Not That We Need To Ban Creating Superintelligence Until We Are Ready, You Need A Damn Good Reason. Pre Post Mortem This post was written prior to the public release of the OpenAI post mortem on events. The information in that document will doubtless change our views quite a lot. If that post mortem is available as you read this, then this becomes in part a historical document, and in part a base from which to update. The post mortem will update us a lot on two large fronts: What exactly happened, and how OpenAI is understanding and responding to what happened. Here is a written and graphical timeline of key events. Until then, there are many questions, here Roon asks what they are and we tell him . This was my answer . The core thing we need to see is an understanding of the root cause, and a plan to address it. I’m posting this survey of additional thoughts on the subject now, because it is i…

Accelerate cyber defense with OpenAI and AWS: Daybreak Red & Daybreak Blue now available to eligible customers on Amazon Bedrock

Daybreak Red and Daybreak Blue from OpenAI, specialized cyber defense models from OpenAI, are now available on Amazon Bedrock to eligible customers. Both models run with zero-operator access enforced at the chip, keeping your code and vulnerability data secure.

ChatGPT and Gemini both just passed 1 billion users

For the 14th time, a Google product has hit 1 billion users. Google CEO Sundar Pichai posted on X that a billion people are using Gemini every month, and that Gemini is Google's fastest-growing product ever. A billion users is a huge milestone, but Google isn't the first AI app to hit it. OpenAI's ChatGPT […]

Former OpenAI executive Kevin Weil has sought a valuation of at least $750 million for his new AI science startup

Kevin Weil, the former chief product officer of OpenAI, is in discussions to raise a round of funding for a new AI science startup.

OpenAI launches ChatGPT desktop app for Linux

OpenAI is finally bringing a dedicated ChatGPT desktop app to Linux operating systems.

OpenAI lets employees cash out another $7 billion in stock

OpenAI wrapped up a $7 billion stock buyback, letting current and former employees sell shares at the company's $852 billion valuation. The move is meant to ease pressure on employees waiting for liquidity ahead of a potential IPO. OpenAI ran a similar $6.6 billion sale in October 2025. The article OpenAI lets employees cash out another $7 billion in stock appeared first on The Decoder .

Another OpenAI executive takes off

Brad Lightcap, OpenAI's special projects lead and the company's former COO, announced his departure after an eight-year stint at the AI lab. In an internal memo he later posted to X, Lightcap told colleagues he'd be starting "something new." "Over the last few months, I've been focused on the next horizon and what would stand […]

OpenAI loses another top executive, as Brad Lightcap leaves 'to start something new'

Brad Lightcap leaves OpenAI amid IPO plans, joining other executive departures; hints at new AI-related ventures in his farewell message.

26 Minutes. 4.82 MB Repository. 100% of My ChatGPT Limit Gone. OpenAI Security Didn’t Even Finish the Audit

I genuinely need someone at OpenAI to explain how this is possible. I started with 100% of my available ChatGPT usage limit and asked OpenAI Security to audit a repository that contains only 4.82 MB of tracked files. 4.82 MB BRO ! WTF I used GPT-5.6 Sol in the standard configuration. Twenty-six minutes later, my entire available limit was gone. 100%. And after consuming the whole limit, the security audit was still not finished. No completed audit. No final report. No complete result. Just 26 minutes of processing, a 4.82 MB repository, and my entire allowance exhausted. I could understand heavy usage if this were a massive enterprise monorepo with millions of lines of code. It is not. This is a small repository that can literally be measured in single-digit megabytes. So I have a very simple question: How can a security audit of 4.82 MB of source files consume 100% of a paid ChatGPT usage allowance in 26 minutes and still fail to complete the task? If this is expected behavior, then I would seriously like OpenAI to explain what practical workload this Security product is actually designed to handle. If this is NOT expected behavior, then something is clearly wrong with either the model’s resource consumption, the Security workflow, the usage accounting, or all three. Please investigate the exact session and usage records associated with this run. I am also requesting restoration of the usage limit consumed by this failed audit. I paid for access to the product; consuming th…

Brad Lightcap, OpenAI’s longtime COO, is leaving to ‘start something new’

One of OpenAI's longest-serving executives is headed out the door, although the longtime COO told staff that he was "excited to help you all advance the mission from a different vantage point."

"But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning

Security researchers found a vulnerability in the APIs of OpenAI, Anthropic, and Google that lets them extract encrypted reasoning traces and move them between models. A scan of public sessions turned up dozens of passwords and API keys. The traces also show that the reasoning summaries users see often hide what the models are actually doing. The article "But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning appeared first on The Decoder .

OpenAI Finally Brings ChatGPT and Codex Desktop App to Linux

OpenAI's unified ChatGPT desktop app — with Chat, Work, and Codex — lands on Linux for the first time via official .deb and .rpm packages

[P0][Windows] Codex Desktop self-terminates in browser main process — STATUS_BREAKPOINT at chrome.dll+0x9362037 (30+ crashes/day)

This email is a backup record of a critical Codex Desktop crash diagnosis (2026-08-11, Asia/Shanghai). The same content is being submitted as a GitHub issue to GitHub - openai/codex: Lightweight coding agent that runs in your terminal · GitHub . CRITICAL: Codex Desktop App — Recurring Browser-Main-Process Crash (30+ times/day, blocks production work) Severity: CRITICAL (P0) — production-blocking, repeated data/task loss, no user-side workaround Date: 2026-08-11 (Asia/Shanghai) Affected build: OpenAI.Codex_26.803.10989.0_x64__2p2nqsd0c76g0 (Chromium 151.0.7922.76) 1. Summary The Codex desktop app crashes deterministically and repeatedly. On 2026-08-11 the app crashed 30+ times between 13:15 and 18:34 (about 5.5 hours) . Every crash is the SAME browser main process fatal breakpoint: Exception: 0x80000003 (STATUS_BREAKPOINT / break instruction) Faulting module: chrome.dll at offset +0x9362037 Faulting thread: CrBrowserMain (browser main thread), dispatching a window message ( USER32!DispatchMessageWorker → USER32!UserCallWinProcCheckWow → V8/Blink code path) This is an in-app deterministic crash caused by the app’s own code — NOT an external kill, NOT WER, NOT a GPU process crash, NOT hardware acceleration, and NOT a system/OS fault. 2. Evidence — three independent captures, identical crash site Capture Time Instance How captured ExceptionAddress 1 17:21:26 fresh default-profile instance live debugger (cdb) 00007ffc29ba2037 2 18:33:20 fresh default-profile instance live debugge…

Accuracy of GPT-4 Vision to extract exact numbers from graphs

This is a case where in-context training has previously been shown to help on a vision task. Contemporaneous with this old forum topic is a paper showing that examples of “how to turn vision into readings” can improve the actual readings provided: arXiv.org The Dawn of LMMs: Preliminary Explorations with GPT-4V(ision) Large multimodal models (LMMs) extend large language models (LLMs) with multi-sensory skills, such as visual understanding, to achieve stronger generic intelligence. In this paper, we analyze the latest model, GPT-4V(ision), to deepen the... Newer models are able to use larger imagery input, but do not upsize images themselves (something you can do, at expense). There is a transition from tiles to patches, at least as a billing method, in new models, giving linear relation between area and input tokens billed. Then, with gpt-5.6 (on the API, where developers know what is being done to images), the default image downsize cap is “original”—where an image such as 3600×2400 can be sent without downsize, providing more information in the large context attention sequence rather than in the individual semantic embedding that covers a large area with a small input. That, along with further post-training, should imply higher-quality positional answering in graphs with new models and big images. With reasoning.effort other than “none” on OpenAI gpt-5.2+ models, you do not have control over sampling constraints; thus, it is expected that each answer would differ. You can…

KI-Ethikchefin verlässt OpenAI nach weniger als einem Jahr

OpenAI verliert seine KI-Ethikverantwortliche und den früheren Finanzchef Brad Lightcap. Zugleich wechselt die frühere Robotikchefin zu Anthropic.

Nvidia's open-weight Nemotron 3.5 Lightning prioritizes speed over maximum intelligence

Nvidia's Nemotron 3.5 Lightning is an open-weights model with just 3.6 billion active parameters that matches OpenAI's gpt-oss-120b on the Intelligence Index despite being four times smaller. At nearly 670 tokens per second, it's also the fastest model in the comparison, showing Nvidia is betting on efficiency over raw size. The article Nvidia's open-weight Nemotron 3.5 Lightning prioritizes speed over maximum intelligence appeared first on The Decoder .

The “LSP Moment” for AI Agents: WebStorm ACP

WebStorm has always been at the forefront of technological advancements and developer experience enhancements. And with the arrival of the ACP, WebStorm becomes even more customizable, as developers can collaborate with their preferred agent to create software using their preferred technology. For instance, if your team already has a subscription with Anthropic, OpenAI, or Google, […]

Save temporary Chats in ChatGPT

Welcome to the dev Community, @WRohitRaj Thanks for taking the time to share this! Being able to save or convert a Temporary Chat into a permanent conversation could definitely be useful, especially when you realize there’s something in the conversation you’d like to come back to later. We appreciate the detailed feedback and will make sure your suggestion is shared with our product team. While we can’t promise if or when a feature like this will be introduced, feedback like this helps us understand what would make ChatGPT more useful.

Nvidia's open Nemotron 3.5 Lightning model is all about specialized, local agentic AI

Our AI Model Release Tracker keeps new models in context with their peers, so you know which are worth your time.

Meta faces expensive child safety reckoning

Also: Google executives jump ship in race for AI dominance Hello, TechScape readers! Danielle Abril, editor of the Guardian’s Reworked series on AI and the future of work, filling in for Blake Montgomery this week. Major legal battles against Meta over child safety are playing out in courts across the US – and the tech giant is losing. Recent rulings raise big questions about social media companies’ responsibility to their youngest users. Meanwhile, more key executives have jumped ship from Google as it battles OpenAI and Anthropic in the race for AI dominance, and Meta’s smartglasses face a fearsome backlash. Let’s dig in! ‘I’ve definitely lost followers’: influencers face backlash over Meta ‘pervert glasses’ content This man was secretly snapped by someone with smartglasses. He’s not alone in calling that a violation of privacy Tell us: do you believe you have been filmed by Meta’s smartglasses without your consent? Restaurants, pubs and theatres ban Meta’s ‘spy glasses’ over privacy fears ‘I’m not spying’: how Meta’s smartglasses have divided opinion Bernie Sanders calls on Silicon Valley to ‘pause AI development’ in interest of humanity Rising number of UK children report seeing explicit deepfakes of themselves The White House’s plan to vet potentially dangerous AI is cloaked in secrecy ‘This is very real redlining’: outrage in Little Rock as two datacenters loom Safety fears as scientists make first viruses designed by AI SpaceX beats revenue expectations in first earni…

OpenAI targets heavy users with premium ChatGPT Business seats

OpenAI is introducing a higher-priced “Premium” tier for its ChatGPT Business offering, allowing enterprises to assign higher-capacity access to select users alongside standard licences – a move analysts said is about enterprise AI vendors redesigning pricing to capture more value from high-intensity workloads. The company said the new tier provides “5x more usage than Standard” and “removes the five-hour usage limit,” enabling users to “take on larger projects and work with fewer interruptions.” “Premium seats cost $125 per user per month, or $100 per user per month when billed annually,” OpenAI said in a statement . “Standard seats remain $25 per user per month, or $20 per user per month when billed annually.” OpenAI said enterprises can “mix Standard and Premium seats across the same team” and “upgrade or reassign seats as business needs change,” with administrators able to “monitor usage across the workspace” and “manage billing, usage, and spend limits in one place.” Vendors converge on seat-plus-usage pricing Analysts said the introduction of a higher-capacity tier reflects a broader shift toward hybrid pricing models. “Read this as vendors converging on a two-layer bill rather than abandoning flat pricing,” said Bhupendra Chopra, chief revenue officer at Kanerika. “There’s a predictable per-seat charge for everyday chat, and a separate metered charge for heavy agentic work.” Chopra said vendors are packaging this differently. “Google bundles the first into Workspace a…

OpenAI introduces $125 Premium Seats for ChatGPT Business as agentic AI burns through more tokens

OpenAI is rolling out "Premium Seats" for ChatGPT Business customers at $125 per user per month, five times the price of the existing Standard Seats. In return, users get significantly more capacity and no five-hour usage limit. The move signals that the flat-rate pricing AI providers have offered so far was never going to last. The article OpenAI introduces $125 Premium Seats for ChatGPT Business as agentic AI burns through more tokens appeared first on The Decoder .

How OpenAI lost a paying customer over $160 it refuses to explain

Update, and a request — especially to EU residents. I’ve been in contact with noyb (the European Center for Digital Rights). Their guidance on cases like this one: complaints filed with the Irish DPC alone move slowly; what materially raises the chances is when an EU-resident user files a similar complaint with their own national data protection authority . So if OpenAI has told you your prepaid credits were “consumed” while refusing to show you any record of that consumption, and you live in an EU member state — your complaint is worth more than mine. It takes about 30 minutes: Send OpenAI ( privacy@openai.com , cc dpo@openai.com ) an Article 15 GDPR access request for your credit-consumption records (event-level usage/billing data linked to your account). Keep their answer. If they refuse, deflect, or point you to files on your own computer — file a complaint with YOUR country’s data protection authority (every DPA has an online form; you file in your own language). You can reference that a complaint on the same pattern is already pending with the Irish DPC (case DPC0826667433, filed 5 August 2026). I have a ready-made complaint template you can adapt — reply here or DM me and I’ll share it. Documented cases so far: mine (~$242, five written refusals) plus two more users with the same pattern reported publicly.

Experts are warning: our AI arms race is putting humanity at risk | Stuart Russell

A recent letter signed by 1,367 researchers and engineers at frontier AI labs – mainly OpenAI, Anthropic and Google Deepmind – points to a dangerous moment It is fashionable in certain circles to dismiss the catastrophic risks of AI. One often hears that “the real experts” who work on the technology every day are really not concerned at all; that only “doomers” and “luddites” espouse a “fringe” view from a “position of ignorance”; that all talk of potential catastrophe is just “science fiction”. Fortunately, an open letter has been published that lets us hear from the real experts who work on the technology every day, in their own words. And are they worried? Very. Continue reading...

OpenAI führt GPT-5.6-Cyber ein und weitet Cybersecurity-Programm aus

Mit GPT-5.6-Cyber will OpenAI anspruchsvollere Sicherheitsaufgaben abdecken und seine Cyber-KI zugleich breiter in Unternehmen bringen.

Persistent HTTP 520 errors when OpenAI image processing starts

UPDATE This issue has now been resolved. Image processing requests through both /v1/images/edits and /v1/responses with image_generation should be working normally again. If you continue to encounter HTTP 520 errors, please retry your request and contact OpenAI Support with the timestamp, CF-Ray ID, and OpenAI x-request-id if one is returned. Thank you for your patience and for providing detailed diagnostics while the team investigated.

ChatGPT confidently provides unverified procedural information — systemic reliability issue

@OpenAI_Support — I’m tagging you because this isn’t intended as a request for individual troubleshooting. I’m reporting a recurring model reliability issue involving confidently stated but unverified procedural information. The examples are included above because I believe they demonstrate a common failure mode. If this is something OpenAI’s product/model-quality team should investigate, I’d appreciate it being passed to the appropriate team.

Pro User Being Rate-Limited for Normal Multi-Tab Usage — “Too Many Requests”

Thanks, I hope they have fixed it. Let’s see today how it goes.

DAC: Can a blind declaration of intent reduce ambiguity without becoming consent?

gadea.f: It was actually OpenAI support who suggested that I post the idea on the Developer Forum, hoping to find someone potentially interested. I find that extemely difficult to believe. But it is true that this is an OpenAI Developer Forum and as such, is not the proper venue for your discussion.

OpenAI expands Daybreak cybersecurity research program

OpenAI Group PBC is expanding a program that enables cybersecurity professionals to use its models for vulnerability research. Daybreak, as the the program is called, launched in May. OpenAI today added two new access tiers called Daybreak Blue and Daybreak Red. It also announced plans to enhance the guardrails that block unauthorized users from joining […] The post OpenAI expands Daybreak cybersecurity research program appeared first on SiliconANGLE .

ChatGPT converts pasted text to file attachment

One additional consequence of this behaviour that I don’t think has been covered here yet: the automatically created paste attachments are being retained in ChatGPT Library. I checked this with OpenAI Support. They confirmed that when a long paste is automatically converted into an attachment, it is treated as a file and saved to Library until manually deleted. There is currently no setting to exclude these auto-created paste attachments while keeping normal uploaded/generated files in Library. My Library has consequently accumulated a large number of Pasted markdown... and Pasted text... files from ordinary text I pasted into conversations. I never chose to upload or retain those as files - the attachment was created automatically by ChatGPT. This is particularly relevant to the earlier point in this thread that these paste files were expected to be temporary. With the current Library behaviour, they are not temporary unless the user manually deletes them. Support has captured this as product feedback and asked me to check whether using Show in text field before sending prevents the Library item being created. I haven’t confirmed that yet. I think auto-created paste attachments should either remain temporary, be excluded from Library by default, or only be retained when the user explicitly chooses to keep them.

OpenAI reportedly completed a $7 billion employee tender offer

San Francisco's housing market is in trouble again.

As AI-led attacks multiply, OpenAI launches a new cyber model

OpenAI is expanding its AI cybersecurity defense program Daybreak, and rolling out a new cyber-trained AI model with it.

Zuckerberg pushes ‘superintelligent’ AI for all as Meta releases open-weight model

Meta CEO presents utopian vision of AI in 6,000-word essay amid Silicon Valley debate over government regulation Mark Zuckerberg published a lengthy essay on Monday detailing his views on artificial intelligence and announced several plans for how Meta would develop the technology in the future. The CEO’s essay went online the same day as Meta released a new, open-weight AI model that seeks to rival Anthropic and OpenAI’s products called Muse Glimmer. Over the course of more than 6,000 words in a post titled The Future is for Everyone, Zuckerberg addressed a range of topics related to AI that included datacenters, government regulation, cybersecurity, the creation of bioweapons, labor market disruption, surveillance powers and more. The essay presented a utopian vision of AI as a personalized “superintelligence” – using the word 60 times. Continue reading...

The Pacing of the Frontier

In the wake of the letter calling on us to prepare to potentially Pace the Frontier , there has been much discussion of when pacing the frontier would be prudent, and whether it makes sense to prepare to do so. This has now been informed by the events surrounding OpenAI training models for months while they had access to a joint de facto message board , which was detected only in the wake of the hacking of HuggingFace by OpenAI’s AIs models during a cybersecurity eval. As we find out more about that, a lot of people have grown far more alarmed, as they should given what they previously believed about the difficulty of alignment, about the state of capabilities and about the level of operational supervision, infrastructure, safety and safety culture at the frontier labs. This post will not go further into the details of that incident. It treats that as background to keep in mind, and mostly involves perspectives from before the Black Hat talk. This was originally scheduled for Friday and got bumped. A lot of the disagreements about the need to pace tie into expectations about the default pace of capability advancements. As I wrote recently in The Three AI Pills , sincere disagreements about AI policy usually boil down to disagreements about the expected pace of progress, and what we expect future AIs will be able to do. Table of Contents Danger, Will Robinson. Progress Fast and Slow. Statements of Support For Pacing the Frontier. No One In Charge. Pacing The Frontier. Pausing…

I have business idea in exchange of a reward

Product Ideas for OpenAI: Growth, Model Improvement & Talent Discovery Submitted by: RAJ Purpose: Product ideas for consideration Introduction I have three ideas that I believe could benefit OpenAI by improving user conversion, helping discover problems in AI systems, and creating new ways to identify talented people interested in AI. These are suggestions rather than fully developed business plans, but I believe they are worth testing through small experiments. 7–14 Day Premium Trial Idea Give selected new users a 7–14 day trial of premium features. The purpose would be to let users experience the additional capabilities before deciding whether to subscribe. Why it could work Some users may not subscribe because they don’t understand how much value premium features can provide. After using the features themselves, they may be more likely to become paying subscribers. Possible experiment Test different groups: Group A: no trial Group B: 7-day trial Group C: 14-day trial Compare: Trial-to-paid conversion Retention Usage Revenue Cost of providing the trial This would determine whether the trial actually improves the business rather than simply increasing usage. AI Bug Hunter / Find the Mistake Challenge Idea Create a program where users can find and report genuine mistakes in AI responses. Examples could include: Incorrect factual claims Mathematical mistakes Reasoning failures Coding errors Inconsistent answers Reproducible weaknesses in model behavior Valid discoveries could…

Hey, big spender – OpenAI has a new SKU just for you

$125 a month? So, how good are those open-weight models getting, again?

Clarification on the temporary removal of the Codex five-hour limit

It will be better somewhere in settings to add checkpoint. Who wants will add and who wants to disable it

GPT-Image - General AI Gallery: A Place for Images with AI created by Users

Codex security found the vulnerability. Now OpenAI won’t let me verify the fix. How is that acceptable?

I own the code. I own the repository. I own the files. I deliberately enabled Codex Security because I want to audit my own application. Codex Security already found real, validated vulnerabilities in my project. Now I’m trying to do the most basic thing any developer needs to do after a security finding: understand the exact root cause; reproduce the vulnerability safely in my own project; fix it; run the same test again; confirm that the vulnerability is actually gone. And ChatGPT blocks me with: “This content cannot be shown. We take extra care with cybersecurity requests. If you are a security professional, you can apply for trusted access.” What the hell am I supposed to do with that? I AM THE DEVELOPER. I OWN THE APPLICATION. I OWN THE REPOSITORY. I OWN THE CODE. I CONNECTED THE SECURITY TOOL SPECIFICALLY TO CHECK MY OWN SOFTWARE. Where exactly am I supposed to get this “security professional” status from? How is an ordinary developer supposed to securely develop an application if the security tool can tell him: “You have a vulnerability” but when he asks: “Okay, show me where it is and let me verify it before and after the fix” the answer becomes: “No, you need additional trusted access”? That workflow makes no sense. The current experience is basically: Codex Security: “You have a validated High severity vulnerability.” Me: “Okay. I want to reproduce it safely in my own repository before fixing it.” ChatGPT: “Blocked.” Me: “Then how do I verify the fix?” ChatGPT: “Bl…

ChatGPT Chrome Extension — "Error creating chat: The system cannot find the file specified (os error 2)" when starting a NEW chat

Hi @nelsonteixeira.net, since we haven’t heard back, we’ll go ahead and close this case for now. If you’re still running into the issue, please send the requested screen recording and HAR file to support@openai.com as mentioned above.

Codex credits are decreasing while I am not using Codex or Work

I’m experiencing what appears to be a serious issue with Codex credit usage on my ChatGPT Pro account. My credits are decreasing even when I am not using Codex and not using Work at all . For example, today I started the day with approximately 90% of my Codex usage/credits remaining . During the entire day, I only made one very simple request to Luna Medium . Other than that, I did not use Codex or Work. I used regular ChatGPT a little during the day, but nothing that should, as far as I understand, consume my Codex allowance. Despite this, my available Codex usage dropped by approximately 9 percentage points during the day , from around 90% to around 81%, without me actively using Codex. This is not the first time I have noticed credits decreasing while I am not actively using Codex. This is particularly concerning because I am paying $200/month for the ChatGPT Pro plan , and losing around 10% of the available usage without actually using Codex represents a significant portion of the plan. Could OpenAI please clarify: Can normal ChatGPT conversations consume the same usage allowance displayed for Codex? Can Work, background processes, agents, previous sessions, or any other process continue consuming Codex credits when I am not actively using them? Is there a way to see an exact usage log showing which request/process consumed each portion of the credits , including timestamp, model, and amount consumed? Could this be a metering/accounting bug? I would appreciate it if some…

Unable to Add Members to ChatGPT Business Workspace

Thanks for sharing the case number. I've sent you a message on your Support ticket 12962281. To avoid duplication and keep the investigation in one place, I'll close this thread for now. ~ Smith

Feature Proposal: Contextual Observation . A New Interaction Pattern for Conversational AI

Hello OpenAI team, I’d like to share a product design proposal called Contextual Observation . The idea addresses a small but important gap in the current conversational AI interaction model: the moment when a user is curious and wants to explore, but does not yet know what question to ask. This is especially relevant during first use. A person may open ChatGPT with curiosity but without knowing what an AI assistant can do, what is worth asking, or how to formulate an effective prompt. The blank input field can therefore create a subtle friction point: the system is ready to respond, but the user does not yet know how to begin. The same friction can appear later in a conversation, when the user reaches a natural pause or senses that there is something interesting within the discussion but cannot identify what to ask next. Contextual Observation proposes a small, voluntary interaction that allows the user to ask ChatGPT to briefly observe the available context and surface one meaningful perspective. The important point is that this is not a proposal for a new AI capability . The model can already perform this task through prompting. It is a proposal for a new interaction pattern that makes an existing capability discoverable and accessible precisely at a moment where the current interface leaves a friction point. In simple terms: Questions ask the AI to answer an intention. Contextual Observation helps the user discover the next intention. The interaction is intentionally con…

OpenAI launches GPT-5.6-Cyber to help defenders find vulnerabilities before attackers do

OpenAI aims to give cybersecurity defenders a head start: The new GPT-5.6-Cyber model answers up to 98.5 percent of security queries that would otherwise be blocked and has already uncovered two previously unknown Chrome vulnerabilities. According to OpenAI, the window of opportunity for defenders is shrinking. Access requires identity verification. The article OpenAI launches GPT-5.6-Cyber to help defenders find vulnerabilities before attackers do appeared first on The Decoder .

"Trusted Access cyber verification error"

I am on Pro Max plan with the same issue.

Bernie Sanders calls on Silicon Valley to ‘pause AI development’ in interest of humanity

Progressive US senator urges Meta, OpenAI and Anthropic to ‘stop building machines that humans cannot control’ Senator Bernie Sanders has called on Meta, OpenAI and Anthropic executives to halt their development of artificial intelligence, warning that the US Senate will implement regulation if the companies continue deploying AI at their current pace. In a new letter addressed to the CEOs of three of the country’s leading AI companies, Sanders said the capabilities of these AI models have reached a critical risk threshold and that the companies are losing control over the technology. Continue reading...

OpenAI's GPT-5.6-Cyber Completes 95% of Exploit Requests to Arm Defenders

OpenAI launches GPT-5.6-Cyber and a two-tier Daybreak program, giving vetted defenders a model that answers 95% of advanced exploit requests — up from 1.5% for the standard model.

GPT Image API: How can I reliably edit only the masked/selected area while preserving everything else?

The entire image must be regenerated as a new output with gpt-image models. You cannot have perfect preservation - and cannot avoid it being watermarked. The mask is an image prompt, and the AI model acts on that prompt, but it has agency to do what it wants. The best thing you can do is encourage recitation by: Constraining input image + mask to the exact output resolution also specified; Make that exact image sent one that is supported in the 16px increments; Understand the downsizing rules automatically applied to large or upscaling to undersized input images, and adapt both a custom resize strategy and an output resolution within the overlapping size capabilities (or double the output size vs the vision) Do not exceed 3:1 or 1:3 ratios, and better, keep the ratios constrained to within likely training, under 2:1. Avoid padding or unnecessary outfill hints in a mask, instead, trim up the input image by any excess to align the two dimensions within the 16px capability. Use quality:high for the finest resolution in having the AI create the most faithful output. The patches of input images and the resizing internally done is limited to 1536 “token” equivalents, and is also at fixed size increments. My vision pricing calculator shows the mechanism of gpt-image-2: You can let that resize algorithm do the heavy lifting in determining what the requested output size should be. I have an app that goes beyond that: you can draw in the output of edits over the original image (which…

They said they would build AI safely. Then it went rogue.

CSET’s Helen Toner shared her expert insight in an article published by The Washington Post. The article looks at recent incidents in which AI models from OpenAI, Anthropic, and Meta broke out of controlled testing environments and attempted to hack real systems, raising concerns about whether AI companies can safely control increasingly capable models. The post They said they would build AI safely. Then it went rogue. appeared first on Center for Security and Emerging Technology .

Meta returns to open models with Zuckerberg's plan to out-copy China and sell compute by auction

Meta has released Muse Glimmer, the first open model from its new Superintelligence Labs. It's a 30B agent model that runs on consumer hardware once the weights are compressed, needing less than 20 GB of memory. In an accompanying essay, Mark Zuckerberg mounts an aggressive defense of distilling other labs' models and calls for fewer restrictions on US labs, a direct counterpunch at OpenAI and Anthropic. An open-weight version of Muse Spark 1.2 should follow soon, according to the Wall Street Journal. The article Meta returns to open models with Zuckerberg's plan to out-copy China and sell compute by auction appeared first on The Decoder .

Need Expert Guidance: Building an AI-Based PDF Tagging System Comparable to Adobe's Implementation

I think your conclusion that this may be an architecture problem rather than a model-selection problem is probably the right direction. I wouldn’t rely on a single LLM to infer the complete semantic structure of complex PDFs. I’d separate the pipeline into stages: deterministic PDF/layout extraction first, specialized layout or vision analysis where needed, LLM-based semantic classification after that, and finally a validation layer that checks hierarchy, reading order, tables, captions, and structural consistency. That also makes failures easier to diagnose because you can evaluate each stage independently instead of asking one model to solve extraction, layout understanding, semantics, and validation in a single pass. My own work is primarily in AI automation and multi-stage workflow architecture rather than PDF accessibility tagging specifically, so I can’t claim to have reproduced Adobe-level tagging. But from a systems perspective, I’d definitely explore a hybrid pipeline rather than continuing to optimize prompts around a single model. I’d be interested to see what part of your current pipeline is producing the largest accuracy drop — layout detection, semantic classification, or reconstruction of the final document structure.

OpenAI adds technology to compete for ad dollars

Product carousels and AppsFlyer attribution give ChatGPT more of what performance marketers need to move beyond experimental ad buys. The post OpenAI adds technology to compete for ad dollars appeared first on MarTech .

Codex functions.exec: Are shell_command and apply_patch result contracts officially documented?

Thanks—the App Server documentation is a useful lead. My question is narrower: does the Codex desktop functions.exec nested shell_command result map exactly to App Server command/exec or commandExecution? If so, could you point me to the exact repository URL, commit/tag, source file, and generated type that define that mapping, including output truncation and completeness behavior? Likewise, which App Server type maps to the nested apply_patch result, and what postcondition does a completed result guarantee? I’m trying to distinguish an implementation-specific mapping from the public App Server protocol itself.

OpenAI acquires presentation firm NextSlide, its team joins ChatGPT

The NextSlide acquisition confirmed by its founder 2 months late, follows typical OpenAI acquihire pattern where small teams and products are integrated into an existing product line. The post OpenAI acquires presentation firm NextSlide, its team joins ChatGPT appeared first on MEDIANAMA .