Updated

Researchers found ways to bypass Anthropic's Claude safeguards intended to prevent misuse for bioweapons research, highlighting the difficulty of distinguishing dangerous biological research from legitimate scientific work when designing AI safety guardrails.