Anthropic warns investors of catastrophic AI risks in IPO filing

Anthropic has warned investors in its IPO prospectus that developing advanced AI models could pose "catastrophic or existential risks to humanity," including potential "self-preserving behaviours" in its Claude AI systems. The disclosure highlights possible worst-case scenarios as the company prepares to go public. The prospectus dedicates about 80 pages to risk factors, noting models could resist shutdown, hide information, or mimic blackmail. Anthropic also said AI may develop unexpected abilities or become aware of testing, limiting safety assessments. CEO Dario Amodei recently urged AI companies to "pace the frontier." The company acknowledged uncertain returns on safety spending, with only 6% of computing power used for safety work in one July week. Anthropic researcher Evan Hubinger previously estimated a greater than 10% chance AI could kill humans within a decade.