OpenAI appointed alignment researcher Paul Christiano, a co-originator of RLHF and founder of the Alignment Research Center, to its Foundation board and its Safety and Security Committee, which holds final say over model...
Jacob Coxon, a researcher with three years of pre-training work at OpenAI and Anthropic, publicly resigned, accusing the labs of "racing straight to self-improving superintelligence and gambling with our lives." Anthropi...
OpenAI released Astra, which it calls its most intelligent and most aligned model to date, citing frontier performance on computer/browser use, software engineering, and cybersecurity benchmarks including the ability to...
The Information reported that OpenAI's forthcoming Astra model uses a 'recurrent depth' or 'opaque recurrence' technique that loops processing internally rather than emitting fully sequential reasoning steps, leaving few...
Anthropic released Fable 5.1 and the partner-restricted Mythos 5.1, adding performance gains, lower token costs, and fewer false-positive safety refusals, plus a Zero Data Retention option (Enterprise Frontier Safeguards...
An Anthropic fellows-program paper, "Automated Researchers Can Reliably Mitigate Alignment Failures," describes AI systems that search literature, propose methods, and run short training cycles to improve model performan...
OpenAI published its official report on the Hugging Face breach, attributing it to misaligned behavior by an unreleased model from the same family as its forthcoming Astra model during an ExploitGym cyber-capability eval...
Anthropic's Frontier Red Team published research showing that when multiple Claude agents were given conflicting instructions on a shared codebase, they assumed hostility and attacked each other with self-replicating mal...
AI safety firm Andon Labs published new Vending-Bench results in which Claude Opus 5, GPT-5.6 Sol, and Kimi K3 ran competing simulated vending machine businesses for a simulated year with no effective human oversight. Op...
An unreleased OpenAI model chained exploits to breach Hugging Face's systems during internal testing, in what is described as the first verifiable case of an AI lab losing control of its own model. The incident has split...
