Swiss AI security lab 0sec has released 0 on GitHub, an open-source harness for running autonomous security investigations against authorized code, binaries, applications, and infrastructure. The system coordinates parallel agents, security tools, sandboxed verification, and proposed fixes through a browser interface, command-line client, or Model Context Protocol server.

The release matters for developers because it packages work that teams often assemble themselves: connecting models to security tools, dividing investigations among agents, reproducing suspected vulnerabilities, preserving evidence, and integrating the results into development workflows. According to the maintainers, 0 has contributed fixes affecting AMD, Qualcomm, Cisco, Arm, Google, Meta, IBM, NVIDIA, Intel, and Red Hat. Its disclosure log also includes Linux kernel CVEs across Bluetooth, NFC, mac802154, MCTP, and mlx5e subsystems.

The documented installer supports Apple Silicon macOS and x64 or ARM64 Linux. Running 0 starts a local engine and opens a browser console without requiring a 0cloud account. Users supply their own model provider, so code and prompts may still pass through that provider under its data-handling terms.

0 browser interface displaying a security investigation
The local browser interface combines chat, findings, evidence, and run management.

What ships in the harness

0 provides a control plane around models and security tools, with each interface addressing a different part of the workflow.