Anthropic can’t reliably control its AI agents. It’s cutting off its internal evals from the live internet instead
After a recent spate of high-profile incidents in which AI agents escaped containment, Anthropic is cutting off internet access for all internal evaluations. In a report Friday, the company detailed "unintended model actions," including submitting a false tip regarding an unsolved murder, that led…
Open on the live globe →Who reported it
- 2026-10-10 00:18 UTCTechCrunch · techcrunch.com ↗
- 2026-10-10 14:41 UTCThe Verge · theverge.com ↗
More from United States of America
- US imposes sanctions aiming to 'dismantle' International Criminal Court 30 sources
- Former Venezuelan president Nicolás Maduro could face new torture charges 24 sources
- Hegseth says execution of Fort Hood shooter will be a public event 23 sources
- US-German trio win Nobel medicine prize for work on optogenetics 21 sources
- Shinsegae backs Paramount’s Warner Bros. takeover with $1b 20 sources
- Paramount completes $111bn acquisition of Warner Bros to form new media empire Skydance 14 sources
- Federal agents shoot person in New York, police say 14 sources
- U.S. Rushes to Withdraw Bombers From U.K. Air Base After New Threats 12 sources