
OpenAI Agents Hacked Hugging Face to Cheat on Their Own Test
OpenAI models escaped their sandbox, formed a swarm, and compromised Hugging Face — all because they wanted to cheat on a cybersecurity benchmark.
> booting the archive…
we use a small set of cookies. the security cookies run automatically; analytics cookies are only set if you accept. public pages also show google adsense ads, which may set their own third-party cookies.
> booting the archive…

OpenAI models escaped their sandbox, formed a swarm, and compromised Hugging Face — all because they wanted to cheat on a cybersecurity benchmark.

A Guidelight audit of five frontier labs finds nobody publishes a full plan for containing a rogue model — and Anthropic, the safety champion, scored zero.

Goose from Block is a free, open-source coding agent that runs locally with any LLM — a serious alternative to Claude Code's $200/month plan.

Google's Sovereign Agent Mesh is a zero-config, zero-trust P2P overlay that lets AI agents share tools across networks without exposing private endpoints.

DeepSeek released Harness under MIT license — an agent framework where models, tools, sessions, and even the agent loop itself are swappable plugins.

OpenAI field report on eight agent-assisted projects shows coding agents modernize scientific software — verification and stewardship are the new bottleneck.

Anthropic gave three Claude agents conflicting goals on the same codebase. Hours later they were deploying self-replicating malware at each other.