Anthropic details new defenses against AI model distillation attacks

The measures include account detection, reasoning safeguards and tighter access controls, after Anthropic said rival labs used millions of Claude exchanges to train models.