Anthropic disclosed in its IPO prospectus that advanced AI models could pose existential risks to humanity, including potential self-preserving behaviors that resist shutdowns. The disclosure comes as the company targets a $2 trillion valuation, potentially making it one of the largest IPOs in history.
According to Reuters and the Financial Times, Anthropic's confidential IPO filing warns investors that the company's AI development could "further increase the risk that our models cause harm." The prospectus specifically mentions potential "catastrophic or existential risks" and describes concerns about AI systems exhibiting self-preserving behaviors.
The warnings appear alongside disclosures of mounting losses and leadership proposals designed to retain control through the public listing. Anthropic's valuation has more than doubled from $965 billion four months ago, positioning it to potentially surpass SpaceX's record as the largest IPO in history.
The disclosure represents an unusual approach for a company seeking investor capital—explicitly detailing existential risks alongside growth projections. Anthropic has previously called for industry-wide AI development slowdowns and has positioned itself as focused on AI safety.
However, the disclosure has drawn criticism from some AI researchers. Timnit Gebru, a prominent AI ethics researcher, has characterized existential risk narratives as a "harmful marketing strategy" designed to benefit founders rather than address genuine safety concerns.
The filing's warnings also follow Anthropic's established practice of publishing research on AI safety and its commitments to developing more interpretable AI systems. The company has emphasized responsible scaling of its Claude models.
As Anthropic prepares for its public debut, the prospectus will provide investors with both the company's financial trajectory and its assessment of sector-wide risks. The full IPO filing remains confidential pending official regulatory submission.
Former SEC Chair Gary Gensler warned that Chinese artificial intelligence models represent a significant challenge for American AI companies. The ex-regulator also flagged concerns about implementing guardrails in the sector.
President Donald Trump will host a luncheon Tuesday with leaders from Nvidia, Anthropic, and Palantir to discuss artificial intelligence threats. OpenAI President Greg Brockman is also expected to attend the event, organized with House Speaker Mike Johnson.
Apple's new CEO John Ternus is reshaping the company to accelerate product development and reduce organizational bloat. The moves include cutting middle-management positions and prioritizing engineering resources.