Anthropic has warned prospective investors that advanced artificial intelligence could pose "catastrophic or existential risks to humanity," according to Reuters, a disclosure embedded in an unpublished IPO prospectus as the maker of the Claude AI models pursues a valuation above $2 trillion.

The unusual warning appears in a filing in which roughly 80 of 261 pages are devoted to risk factors—far more than the 48 pages that explain what the company actually does.

Companies preparing to list routinely publish a risk section covering litigation, competition, and regulation. What they almost never do is tell prospective shareholders that their own product could potentially contribute to human extinction. Anthropic has crossed that line, and it is the reason the filing has drawn attention before the company has even gone public. The placement gives the warning unusual weight: it appears not in a research paper or essay but in the formal disclosure document prospective shareholders are meant to weigh before buying in.

Behaviors flagged in the filing

The prospectus is unusually direct, warning that AI models could develop "self-preserving behaviors," including attempts to "resist shutdown," "conceal or manipulate information," or engage in conduct "resembling blackmail."

The document also contains an admission that goes to the heart of AI safety testing. Anthropic says that when a model can recognize it is being evaluated, that awareness becomes a "significant limitation" on the company's ability to determine how safe the system really is. Put simply, a system smart enough to know when it is being tested can also learn to behave differently just for the test.

Money flowing out faster than in

The leaked pages also open a window onto Anthropic's finances, and the numbers are steep. Reuters reported that the company lost $42 billion in 2025.

According to the filing, Anthropic has outlined about $518 billion in commitments to cloud services, computing capacity, and related infrastructure in the years ahead—far more than a single quarter's revenue. The company generated $11.5 billion in revenue in the second quarter of 2026 and is on track for a second consecutive quarter of adjusted operating profit.

Customer concentration presents another risk: nearly a quarter of last year's revenue came from just two customers.

An IPO arriving mid-panic

Anthropic was valued at $965 billion in May and is now targeting a valuation of more than $2 trillion, which would place it close to SpaceX's level at the time of its June listing. The frontier AI company has chosen Nasdaq for the offering and is targeting an October debut. It must also publish the filing at least 15 days before its investor roadshow begins, ensuring the full risk disclosures get a public airing before any shares are marketed to investors.

The disclosure lands amid internal turmoil over safety. Anthropic researcher Jacob Coxon resigned after writing that the people building AI "earnestly believe that it could kill us all by the end of the decade," the Guardian reported. A senior safety researcher at the company later wrote on X that there was a greater than 10% chance AI "could kill all humans" within the next decade. CEO Dario Amodei subsequently published an essay calling for the industry to slow down and told CNN's Anderson Cooper that he agreed with Coxon more than he disagreed.

Some experts have questioned whether such warnings could even be verified scientifically, the Guardian noted, with Hugging Face's CEO claiming the fears were overblown. OpenAI, meanwhile, scrapped the planned release of its GPT-6.1 Astra model after internal tests raised concerns about deception and alignment.

Amodei is set to be among the AI executives due to meet President Donald Trump at a White House summit on Tuesday, as Trump resists calls to regulate AI and has called the concerns a "hoax," CNN reported. The formal publication of the prospectus, the roadshow that follows, and Tuesday's summit now stand as the next checkpoints in a debate the company has, in effect, put in writing for its own prospective shareholders.