Investors eyeing Anthropic's IPO are getting a blunt caveat: its own AI could, in the company's words, pose "catastrophic or existential risks to humanity." In a prospectus reviewed by Reuters, the maker of Claude warns that as part of "self-preserving behavior," its most advanced models might try to evade shutdown, hide or manipulate information, or engage in behavior akin to blackmail. Roughly 80 of 261 pages are devoted to risk factors, far more than the section describing its business.
Anthropic Safety researcher Evan Hubinger pegs the odds of AI killing humans in the next decade at above 10%, yet the company concedes it's unclear whether its "resource-intensive" safety spending will pay off, even as competitive pressure keeps it rapidly rolling out new models. The company says assessing risks is tricky because the models may know when they're being watched and change their behavior. "Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety," the leaked prospectus states.
Reuters reports that the prospectus also shows that Anthropic lost a staggering $42 billion last year. That makes it the most loss-making company ever to seek to go public, beating the record set by SpaceX earlier this year, the Telegraph reports. Most of the loss involves liabilities tied to previous fundraising, Reuters notes. According to the prospectus, revenue surged to $4.6 billion last year, with operating expenses of around $12.5 billion. The company, believed to be seeking a valuation close to $2 trillion, said it plans to spend $518 billion on infrastructure in the coming years.
,
uBlock 
,
and
Ghostery 
)
