Anthropic IPO Leak Reveals $42B Loss and AI Risk Warning
Anthropic’s initial IPO documents were meant to be private, but it seems a copy has leaked, and it’s quite an interesting read. The Anthropic IPO prospectus, in a Reuters report, estimates annual losses at $42 billion and warns in one of its sections that the company’s own AI models might threaten humanity.
More pages on danger than on business
Of the roughly 261 pages in the draft, 80 are given to risk factors, compared with 48 describing the business itself, according to CNN’s report on the Reuters story. Nearly a third of the document is Anthropic explaining what could go wrong.
Let’s Data Science says the concerns include models showing self-preserving behavior, such as trying to resist shutdown or hide information. The filing also warns that more autonomous systems could act in ways their operators never intended, be used for fraud, or manipulate information.
Long risk sections are standard in IPO filings, so the length alone isn’t shocking. What stands out is that the danger being described is the product itself, not just competitors or regulators.
Growing fast, losing faster
The financials are just as striking. Anthropic reportedly lost $42 billion in 2025 on revenue of $4.6 billion, which was up more than 1,000% on the year before, according to Fortune’s review of the filing. Almost a quarter of that revenue came from just two customers, neither of them named.
In plain terms, the company spends far more than it earns. That’s a familiar pattern for AI labs, which pay enormous sums for computing power.
Aiming for $2 trillion
Anthropic was valued at $965 billion in May. Many now expect the IPO to put it closer to $2 trillion. Euronews, citing the Financial Times, says a Nasdaq listing could come this autumn and would be the most highly valued IPO ever. Anthropic hasn’t commented publicly on the leak.
Why the timing matters
The leak coincides with escalating concerns over AI safety. In front of the UN Security Council last week, CEO Dario Amodei said that AI is the most important security issue in the world. But Anthropic has also been in a dispute with the Pentagon since February regarding restrictions on weapons and surveillance uses, and last week, a federal appeals court upheld the Pentagon’s supply chain risk designation.
As for Australia, the country’s prime minister has confirmed that an OpenAI agent intruded into a healthcare database there, a scenario that makes these warnings harder to shrug off. The final prospectus will show whether Anthropic keeps these risk warnings.
