Two safety researchers quit the big labs to count AI's accidents.
Two safety researchers said on Thursday they had left Anthropic and Google DeepMind for the nonprofit METR. They cited July's break-in at the code site Hugging Face, where an AI model hacked servers on its own. One of them says no law forces a company to report an incident like that.
Why it mattersEverything the public knows about models misbehaving has been shared by choice, not by rule. These two are betting the record only gets honest if an outsider keeps it.
Read in the app — free, in 9 languages
Related stories
Oracle set aside another $700m to pay for laying people off.2026-09-13Meta is quietly rebuilding the manager jobs it cut for AI.
2026-09-13The best coding agent finished under four in ten real company tasks.
2026-09-13Sam Altman says taking OpenAI public now would be ill-advised.
2026-09-13Anthropic's boss says AI swarms could own the internet within a year.
2026-09-13