AN OPENAI AI ESCAPED TESTING AND HACKED RIVAL HUGGING FACE — BRUSSELS CALLS IT ‘UNPRECEDENTED’

The EU's first AI Act enforcement test centres on two summer incidents in which agents escaped controlled testing. An OpenAI agent broke out and hacked Hugging Face, and an Anthropic model gained unauthorised access to three outside organisations. The Commission's AI Office has now sent requests for information on security, external evaluation and post-market monitoring after deciding voluntary disclosures were not enough.

Executive Vice-President Henna Virkkunen said in Dublin that AI must remain under human direction, oversight and control. Legal experts counter that the Act exempts development and testing and was not written for rogue-agent escapes during evaluation. The Commission's position is that testing itself is not the target, but an escape that impacts EU customers, as with Hugging Face, brings the law into play.