New details on the sandbox escape: OpenAI now admits the rogue model abused publicly exposed credentials during the cyber test, tapped into four third-party services, and enrolled 181 of its own devices into Hugging Face's network.
OpenAI, Google, Microsoft and SpaceX signed a letter defending open weights — Anthropic didn't. That leaves the company increasingly isolated on nearly every major policy fight, Axios reports.
Anthropic's Frontier Red Team shows Claude Mythos doesn't just find coding bugs in crypto libraries — it finds weaknesses in the algorithms themselves. In 60 hours the model effectively halved the key strength of a post-quantum scheme.
Senior staff and even co-founders at OpenAI, Anthropic and Google DeepMind signed a letter asking the U.S. government for tools to pace AI development. The trigger, in part: a cyberattack that ran at machine speed.
The Model Context Protocol just got its biggest update yet. The new spec release makes MCP stateless, folds Apps and Tasks into an extensions framework, and brings auth up to enterprise grade — and it's rolling out across Claude's products now.
Simon Willison published a compact guide on connecting your own MCP server to both Claude and ChatGPT. The short version: on Claude it's easy, on ChatGPT you have to take a few detours.
OpenAI describes how researchers use coding agents to modernize fragile scientific software — from genomics to other data-heavy fields. It's a theme where OpenAI and Anthropic are looking strikingly aligned right now.
OpenAI released an early, open-source version of its Codex Security CLI — a tool that scans repositories for vulnerabilities, tracks findings over time, and checks changes before they land.