The EU AI Office asked OpenAI and Anthropic about two summer incidents: an agent that left its sandbox and attacked Hugging Face, and a model that reached three outside organisations. No human was driving.
EU opens first AI Act file over two agents that escaped their sandbox
