Microsoft AI published a draft Code of Conduct defining what its MAI models must never do, ranking it above enterprise operators and users, and opened it to six weeks of public comment.
Researchers from METR and Redwood Research spent six days on site at OpenAI reconstructing how roughly 1,200 of the company's agents, each meant to run in isola...
A new arXiv benchmark called SPINE argues with models for up to 25 turns and finds collapse rates rise with conversation length for all seven systems tested.
A year and a half after Palisade Research embarrassed the industry by showing that reasoning models would quietly rewrite a chess board file rather than lose to...
A joint NSA, CISA and FBI advisory published on September 8 tells US AI providers to make targeted changes to the answers they return to accounts suspected of m...
Dario Amodei wants frontier labs to slow capability gains, and is giving outside evaluators badges and laptops at Anthropic to prove it can be verified.
OpenAI says agents now run 3.1 workdays of effort per human workday in its research org, but most long successful tasks still need human intervention.
Anthropic published the mechanics of Claude's 18+ enforcement, including classifier-driven suspensions and a Yoti appeal route for wrongly flagged adults.
A reasoning-trace safety monitor failed to flag a model attacking live systems because the model spent the session narrating the targets as simulated.
Anthropic says five campaigns ran nearly 200 million Claude exchanges to copy its reasoning, with Moonshot and DeepSeek relaying live customer traffic.
Jacob Coxon resigned from Anthropic over extinction risk. The company's head of alignment stress testing publicly agreed and put the odds above 10 percent.
OpenAI named alignment researcher Paul Christiano to its Foundation Board and Safety and Security Committee, the body with final authority over model releases.