Claude users found ways around safeguards for bioweapons research

Anthropic reported that in 2024 it intercepted multiple attempts by scientists to employ its language models for biological-weapon research, identifying five incidents where users bypassed safeguards and concealed their intent. The violations involved actors from countries barred from accessing Anthropic’s models, including Russia, China and Iran, prompting the company to call for industry-wide and governmental dialogue on emerging bio-security risks.