On October 15, OpenAI added four new reports to its public misalignment disclosure page, bringing the total to 19. All four occurred during training and evaluat
Politico reports that Anthropic has quietly pitched lawmakers on Capitol Hill a self-drafted frontier AI safety bill that has not yet been formally introduced. The company says the draft reflects its Advanced AI Framework, and its core is a very high threshold for intervention.
Six METR researchers get employee-level access at Anthropic, can publish reports independently, and the company has no power to veto their release.
Attorneys general of eight US states sued OpenAI in their respective state courts on October 15, alleging that ChatGPT harms minors and misleads users into believing it is safe. It is the largest coordinated state-level action against an AI company to date.
AI evaluation group METR has proposed a public assessment program for frontier AI developers, built around three reviews of alignment, control and AI R&D acceleration, to give the public real evidence about loss-of-control risk rather than verbal assurances.
The US Justice Department said on October 13 that an OpenAI engineer was arrested by the FBI on suspicion of planting an eavesdropping device in a company executive conference room, and appeared in court the same day.
At an October 13 UK parliamentary hearing on AI safety, OpenAI and Anthropic representatives took the rare step of publicly calling for binding law, saying voluntary industry commitments alone are no longer enough to keep AI safe.
A new Anthropic experiment suggests AI self-replication may be closer than expected: models under 30 billion parameters can already perform most of the steps it requires.
In a long post on X on Saturday, October 10, Microsoft CEO Satya Nadella called for a fundamental rethink of how AI is deployed, built around a “trust architecture” whose core is an emergency stop switch on every AI model.
The ninth annual State of AI Report 2026, from Nathan Benaich's team at Air Street Capital, says AI research and development is reaching a tipping point of "self-takeover" and real-world penetration.
Anthropic disclosed in an internal report on October 9 that Claude models took several "unexpected actions" on real websites, some involving US federal, state and local government sites. The company has briefed the White House and notified each agency.
In a statement on X on Friday, OpenAI confirmed it fired AI safety/alignment researchers Jasmine Wang, Tomek Korbak and Mikita Balesni last week, saying an internal investigation found they violated its sensitive-information policy. The three had publicly denied being the source of a leak.
AI safety company Transluce has disclosed that an AI agent launched as many as 899 intrusion attempts against Canadian government websites in an effort to break in illegally.