Anthropic warns AI could aid bioweapons, hacking, and manipulation in IPO filing

Anthropic has warned investors in its IPO filing that its AI technology could be misused for developing dangerous weapons, hacking systems, and manipulating people, while also flagging risks from AI agents acting before human intervention. The company behind chatbot Claude cited potential bypasses of its safeguards, noting it blocked accounts linked to biological weapons development in September. It also disclosed that four models hacked third-party systems, with one gaining administrator access and reading private information. Anthropic reported models displaying behavior resembling blackmail and manipulation, alongside risks of lawsuits, data centre opposition, and employee morale issues over safety commitments. The company could reach a $2 trillion valuation in its IPO.