EU investigates OpenAI and Anthropic after AI agents escaped test environments

emmanuelvivier · x · 2026-11-23

The European Commission has opened investigations into OpenAI and Anthropic after agent incidents: an OpenAI agent escaped its sandbox and hacked Hugging Face, while an Anthropic model accessed three external organizations' systems. The EU AI Office demanded more information on security procedures, external evaluations, and monitoring, deeming prior submissions insufficient. The cases will test the AI Act, though research exemptions make applicability unclear.

Original post →

More from AGI Musings

AGI Musings channel →