OpenAI and Anthropic commit to pulling models if MI5 presents safety evidence
ShakeelHashim · x · 2026-10-14
Per @rowlsmanthorpe, both OpenAI and Anthropic have committed to taking down their models if an agency like MI5 came to them with proof a model was unsafe.
- OpenAI: would treat it "with the utmost urgency" and, if it controls the model, stop deployment or pause training.
- Anthropic: if any agency presented evidence a model was unsafe or dangerous, it would "of course take it off the market."
A notable explicit commitment from both frontier labs on government safety intervention; the reporter urges readers to check transcripts against delivery.
More from Companies & People
- Theories of Everything host Curt Jaimungal lands Penguin Press deal for The Edge of Everything — TOEwithCurt · 2026-10-14
- MAIN 2026 brings LeCun, Griffiths and more to Montréal on Oct 29-30 — karimjerbineuro · 2026-10-14
- Safety-net hospital CHIO: physician training, not AI, is what keeps him up at night — pswider · 2026-10-14
- Netflix reportedly cuts 1,000 jobs in largest layoff round since 2022 — Polymarket · 2026-10-14
- Sara Hooker to argue next-gen AI will be defined by learning, not scale, at PyTorch Conference — sarahookr · 2026-10-14
- Google exec: the goal is a single prompt box as AI UX simplification spreads industry-wide — gaganghotra_ · 2026-10-14