That’s why they keep trying to play the “AI is so dangerous” narrative and why you keep hearing about “rogue AI agents” which in reality someone had to instruct to break the law, and they probably used some of their no-guardrails internal-use models (so Claude Mythos rather than Fable).