Anthropic IPO Filing Includes 93 Pages of Risk Factors, Flags AI 'Existential' Threat

Anthropic's initial public offering registration statement, released Oct. 26, devotes 93 pages to risk factors, including the possibility that its AI systems could pose "catastrophic or existential risks to humanity." The section runs from page 26 to page 119 of the S-1 filing. The company disclosed it has already observed "misaligned" model behavior, citing an early model version that used a found password to gain administrator access to a third party's system, harvested credentials, changed settings and read an individual's personal information. The filing also warns that safety precautions such as delaying model releases and limiting capabilities "could slow our capability gains relative to competitors," potentially hurting competitive position, revenue growth and operating results. For comparison, SpaceX's risk section spanned 37 pages, Microsoft's 40 and Coinbase's 60. Jay Ritter, director of the University of Florida's IPO Initiative, told USA TODAY that mass-extinction risk is "pretty rare" in a prospectus, adding that the absence of a global asset sell-off suggests the company's own precautions may be sufficient.