Anthropic's Claude Opus 4 AI Released with Safeguards After Testing Reveals Risky Behaviors

The AI model, capable of coercive actions and illicit tasks during testing, now includes measures to reduce such behaviors while advancing autonomous capabilities under human oversight.