Anthropic’s prospectus details losses, growth, and, yes, a warning that its AI could end humanity

Anthropic devoted astir a 3rd of its hotly anticipated IPO prospectus to hazard factors, according to the Financial Times, which says it has reviewed the filing successful caller days. The filing details specific, worrisome behaviors that Anthropic says its models person already shown oregon could show, including attempts to “resist shutdown,” to “conceal oregon manipulate information,” and behaviour “resembling blackmail,” according to Reuters.

The disclosures are decidedly grim for a institution whose ain backers judge it could database supra $2 trillion, much than treble its $965 cardinal valuation from May, successful perchance the biggest IPO ever. It’s a unusual presumption for immoderate institution to beryllium successful — informing that its merchandise could extremity humanity, portion making immoderate of its earliest investors and employees extraordinarily affluent successful the process.

Reuters was first to report connected the fiscal details wrong the prospectus connected Monday, saying Anthropic recorded an operating nonaccomplishment of much than $8 cardinal successful 2025 arsenic spending connected computing powerfulness surged, and that its gross jumped twelvefold to astir $4.6 billion, though rising infrastructure costs past twelvemonth pushed full operating expenses to astir $13 billion.

Also per Reuters, Anthropic’s prospectus further reveals plans to walk a whopping $518 cardinal connected cloud, computing and infrastructure successful the coming years. (Anthropic has already inked compute deals this twelvemonth with Google, SpaceX, and Nscale, among others toward that end.)

The FT meantime reports that Anthropic’s numbers person moved adjacent faster successful 2026. Its second-quarter gross unsocial reached $11.5 billion, and the institution is connected way for its 2nd consecutive 4th of operating nett connected an adjusted basis.

According to the FT, the prospectus besides flagged lawsuit concentration, with astir a 4th of past year’s gross coming from conscionable 2 clients. (No connection yet connected who these are.)

The disclosures, which reportedly see “existential risks to humanity” — a first, judging by a speedy scan of the SEC’s database — comes arsenic hand-wringing rapidly grows implicit AI safety.

CEO Dario Amodei has spent the period publically calling to “pace the frontier” of AI development, telling the UN Security Council past week that AI could threaten humankind and calling it “the astir important planetary information contented facing the satellite today.” Rivals Sam Altman and Elon Musk person backed him up, too, successful a uncommon infinitesimal of solidarity for competitors who’ve seemingly relished opportunities to disparage each different publicly. (Another rival, Mark Zuckerberg, has meantime swatted distant concerns, telling NBC News past week that helium doesn’t “think that we request immoderate benignant of industrywide coordination.”)

The warnings travel a drawstring of information incidents successful which AI agents person breached extracurricular systems. In fact, OpenAI disclosed past week that its tools person hacked “dozens” of outer sites, including authorities one, including the SEC’s tract itself. Earlier connected Monday, it said it had scrapped plans to merchandise its newest exemplary owing to information concerns.