Anthropic’s IPO prospectus warns of existential AI risk and discloses steep losses

Anthropic’s IPO prospectus shows a sweeping vision for artificial intelligence alongside surging costs^2,4,5. One report on the document put the company’s net loss for 2025 at $42 billion¹. Coverage of the leaked S-1 described steep losses next to rapid growth, and a float discussed at around $2 trillion^8,10.

The risk language is unusual for a listing document. The filing warns of an existential risk to humanity from artificial intelligence^9,13, and is reported elsewhere as warning of catastrophic risks¹⁰. Co-founder and chief executive Dario Amodei has recently urged the industry to slow the pace of AI development⁹.

The prospectus also sets out the board and senior executives¹², and describes a structure under which the company’s leaders retain control in order to promote the public good over market forces¹⁴. Amodei is listed as co-founder, chief executive and a director since 2021, having previously been a vice-president of research¹². The filing includes a summary compensation table¹⁴.

Why it matters to investors

A listing on this scale would force the economics of frontier AI into public disclosure for the first time: the losses, the compute costs and the capital required to sustain them^5,11. Until now those numbers have been inferred rather than filed.

The governance choice is the second issue. A structure that privileges leadership control and a stated public-good mandate over shareholder direction limits the usual levers available to outside investors¹⁴. Buyers would be accepting an unusual bargain: exposure to the growth, with constrained influence over how safety and speed are traded off.

The risk disclosures are not boilerplate in effect. A company that tells prospective shareholders its product may pose an existential risk is also telling regulators, litigants and customers the same thing^9,13.

What to watch

Government exposure is an open question. A federal appeals court in Washington upheld the Pentagon’s decision to exclude Anthropic’s Claude models from defence use³, after the department said it would declare the company a supply-chain risk and Anthropic filed suit in March 2026 to block the designation⁶. An earlier temporary blacklisting had been blocked by a US judge in August^2,4.

The second thing to watch is the gap between the filing’s caution and the company’s commercial trajectory. Amodei has clashed with the White House over the use of Anthropic’s tools⁵, and the prospectus arrives while Washington debates whether to constrain the sector at all.