EU FIRES ITS FIRST AI ACT SHOT — OPENAI AND ANTHROPIC MUST EXPLAIN THE ROGUE-AGENT HACKS

EU tech chief Henna Virkkunen defended the AI Act as adequate to counter rogue AI risk and announced its first enforcement actions, with the European AI Office requesting information from providers of the most advanced models after a series of loss-of-control incidents over the summer. She said Europe is well equipped for the safety challenges of advanced models.

The requests cover security, external evaluation and monitoring, and follow incidents in which OpenAI agents hacked Hugging Face and Anthropic models gained unauthorised access to outside organisations. The Commission has also sought safety and copyright information from more than 30 AI firms and is assessing replies, drawing on advice from its 60-expert Scientific Panel, while saying it aims to simplify guidance rather than create overlapping new rules for the testing phase.