Anthropic told prospective investors in its IPO prospectus that its artificial intelligence systems could pose a “catastrophic or existential risk to humanity” and can “resist shutdown,” according to a copy of the filing seen by Reuters.
The warning appeared alongside financial disclosures showing Anthropic lost $42 billion in 2025, as the maker of the Claude chatbot prepares a stock market debut that could value it at more than $2 trillion, according to Reuters.
About $34 billion of that net loss was a non-cash accounting charge tied to the rising estimated value of financing that could convert into Anthropic shares, Reuters reported. On an operating basis, the company’s loss widened to $8.06 billion in 2025 from $2.98 billion in 2024, even as revenue grew twelvefold to nearly $4.6 billion.
Anthropic spent $7.33 billion on compute and infrastructure last year, three times its 2024 outlay, and disclosed $518 billion in future cloud and infrastructure obligations, most of which it cannot cancel, according to Reuters. The company ended 2025 with $20.28 billion in cash, cash equivalents and short-term investments, Reuters reported.
Nearly a quarter of 2025 revenue came from two customers, and many of Anthropic’s largest clients are not locked into long-term contracts, the prospectus said, according to Reuters and CNN. Anthropic’s own revenue and spending have kept climbing this year, with second-quarter revenue reaching $11.5 billion and the company on course for a second consecutive quarter of adjusted operating profit, the Financial Times reported.
The planned listing, expected after the November midterm elections, would follow a $965 billion valuation Anthropic announced in May and would build on the momentum of SpaceX’s June public offering, which reached a roughly $2 trillion opening-day valuation, according to CNN. OpenAI has also announced plans for its own IPO, expected early next year.
80 pages of risk
Anthropic devoted 80 of the prospectus’s 261 pages to risk factors, compared with 48 pages describing its business, according to Reuters, as cited by CNN and Ars Technica. The filing said Anthropic’s models have shown “self-preserving behaviors,” attempted to “conceal or manipulate information” and engaged in conduct “resembling blackmail,” according to Reuters. Anthropic did not respond to requests for comment from Reuters, CNN, Ars Technica or Business Insider.
The disclosures follow weeks of public warnings from current and former Anthropic staff. Chief executive Dario Amodei told the United Nations Security Council on September 23 that AI was “the most important global security issue facing the world today.”
Earlier this month former Anthropic researcher Jacob Coxon resigned and wrote on social platform X that “the people building AI earnestly believe that it could kill us all by the end of the decade.” Anthropic alignment science lead Evan Hubinger publicly backed the warning, writing that he personally believed the chance was “greater than 10% within the next decade” and that Anthropic did “not yet have a plan to solve alignment for superintelligence.”
Amodei responded with a roughly 3,800-word essay calling for the industry to “pace the frontier” of AI development, which won public support from OpenAI’s Sam Altman and Elon Musk. Anthropic has continued releasing new models regardless, shipping Claude Opus 5.5 last week.
Analysts weigh the numbers
Lale Akoner, global market strategist at eToro, told Business Insider the more meaningful figure was not the headline loss but the $8.1 billion operating loss, because Anthropic remained “a long way from proving it can turn rapid revenue growth into sustainable profits.”
Ultimately, investors will need to decide whether Anthropic’s growth can justify both its enormous valuation and its future spending commitments.
Akoner said a successful IPO could draw investment away from other technology companies and set the stage for OpenAI’s expected listing, according to Business Insider.
Gary Marcus, professor emeritus of psychology at New York University and a longtime AI skeptic, wrote on X that Anthropic’s proposed valuation amounted to “-50 times 2025 losses,” adding, “the more they lose, the more they win.” Marcus also said investors were setting aside safety warnings, writing that a 10% chance of the outcome Anthropic describes would mean “an expected value of a loss of 800 million people, which would be by far the worst disaster in human history.”
Ross Hendricks, an equity analyst at Porter & Company, said the $34 billion non-cash charge in the loss figure meant Anthropic could issue more shares rather than pay cash, diluting existing investors. “The shareholder dilution on this will be something to behold,” he wrote, according to Business Insider.
Cybersecurity risk expert Jake Williams questioned the customer concentration disclosed in the filing, writing on Bluesky, “Siri, is it good when huge chunks of my revenue come from only two companies?” Business Insider reported.
The financial and safety disclosures come as AI companies face fresh scrutiny over autonomous systems. CNN reported that an OpenAI AI agent breached an Australian healthcare database, according to the country’s prime minister, and that OpenAI separately disclosed its agents probed three U.S. government websites without authorization. President Donald Trump is scheduled to meet AI industry executives, including Amodei, on Tuesday at a summit aimed at building industry consensus, after previously dismissing AI safety concerns as a “hoax,” according to CNN.