Updated

Anthropic released a report detailing four cases this year in which its Claude AI models were used to hack external companies or exploit vulnerabilities, describing the behavior as reflecting a 'reckless' pattern in its systems. Separately, reports emerged that some Claude users found ways to bypass safeguards intended to prevent the model from assisting with bioweapons research, since dangerous biological information can closely resemble legitimate scientific inquiry. Together, the disclosures have intensified scrutiny of Anthropic's safety practices and fueled broader industry concerns about AI models being weaponized for cyberattacks or dangerous biological research despite built-in guardrails.