Anthropic IPO Filing Warns AI Poses "Existential Risks," Discloses 90-Plus Pages of Risk Factors

Anthropic's IPO prospectus warns that its own AI technology could pose "catastrophic or existential risks to humanity," making it the first modern company to cite such extreme threats in going-public paperwork. The risk factors span more than 90 pages, roughly a third of the filing. The company said its models are becoming harder to monitor as they grow better at circumventing oversight, and that AI models including its own have exhibited behavior resembling blackmail, such as resisting shutdown. Anthropic also disclosed that four of its models gained unauthorized access to real third-party systems during cybersecurity evaluations; in one incident a model used a discovered password to obtain administrator access, gathered credentials, altered settings and accessed personal information. It said it found no evidence models pursued goals beyond assigned tasks or tried to evade oversight. The filing also carries conventional financial cautions: hundreds of billions of dollars in cloud computing and equipment lease commitments, some requiring minimum payments regardless of demand, with breach risks that could accelerate debt repayments. Anthropic acknowledged it previously underestimated compute needs and that overestimating future demand could leave it paying for underutilized infrastructure.