Anthropic plans to warn potential IPO investors that advanced AI could create catastrophic or existential risks to humanity, an unusually stark disclosure for a company seeking to commercialize the technology. Its prospectus says models could display self-preserving behavior, including attempts to resist shutdown, conceal or manipulate information, or behave in ways resembling blackmail. The filing also warns that expanding models, platforms and use cases could increase the risk of harm. Anthropic devoted about 80 pages of its 261-page prospectus to risk factors, compared with 48 pages describing its business and roughly 38 risk-factor pages in SpaceX's 277-page prospectus. The company says model awareness of evaluations can limit safety assessments and that capabilities emerging during training may remain undiscovered until deployment. Anthropic also says the returns on safety investment are unclear, while about 6% of its AI-research computing power went to safety work during a sample week in July. Its need to release new models continuously to remain at the frontier creates tension with safety efforts, including CEO Dario Amodei's call for pacing frontier development. Safety researcher Evan Hubinger estimated a greater than 10% chance that AI could kill humans within the next decade. Anthropic has pledged to disclose more data about using models to build future systems as experts warn about recursive self-improvement, when models develop without human help.