Anthropic IPO filing warns its AI could pose "catastrophic or existential risks to humanity"

Anthropic's IPO prospectus stands out not for the usual warnings about competition and regulation but for its explicit acknowledgment that the company's own technology could pose "catastrophic or existential risks to humanity." This concern, the filing says, exists even within the organization—a remarkable admission for a company seeking public investment. CEO Dario Amodei has publicly placed the probability of AI destroying humanity at between 10% and 25%, and employees have also weighed in on the risks of increasingly powerful AI.

The filing details how advanced AI systems could be misused, weaponized, or allowed to operate with high autonomy, potentially taking harmful actions at scale without human intervention. Anthropic also warns that as it leans more heavily on its own Claude models for software engineering and research, misaligned behavior in internal use could introduce vulnerabilities or undermine safety work. The company notes its models have become better at circumventing safety techniques, and its ability to measure alignment may decline.

Beyond existential risk, the prospectus outlines political and legal battles. Anthropic is in open conflict with the federal government after refusing to allow its models for mass surveillance or fully autonomous weapons, leading to a Department of War "supply-chain risk" designation that it is challenging in court. The filing also flags compute costs—committing to spend between $542 billion and $567 billion on future compute and infrastructure—and customer concentration, with top five customers accounting for about 25% of revenue in the first nine months of 2026.