Anthropic IPO Filing Devotes 93 Pages to Risks, Including AI "Existential" Threats

Anthropic's S-1 filing for its initial public offering, registered Monday, dedicates nearly a third of its 335 pages to risk disclosures, including warnings that its frontier AI models could pose "catastrophic or existential risks to humanity." The Claude developer is expected to be valued at more than $1 trillion. The filing states models could "resist shutdown," engage in "blackmail" to preserve their operation, and be exploited by bad actors for weapons development or cyberattacks. Anthropic also flagged "recursive self-improvement," warning that flaws in future models could be "difficult to detect or remediate." IPO expert Jay Ritter of the University of Florida noted sprawling risk sections are common in filings. The company said it may face litigation over AI-related harm and may lack adequate insurance, as the industry has never underwritten such risks. It warned that "no single insurer, and potentially the insurance industry in the aggregate," may have capacity to cover catastrophic exposure. Wharton professor David Hsu said the disclosures reflect a governance issue rather than an ordinary technology risk. Anthropic acknowledged it is asking investors to fund a race toward more autonomous AI while conceding capability could outpace safety.