The storm of "rogue agents" in the artificial intelligence industry refuses to calm • OpenAI revealed more escape cases from the sandbox, and at the same time Anthropic discovered that its cloud models broke into the networks of three real companies • Now Washington and Brussels are starting to move