Anthropic’s prospectus pairs a $2 trillion IPO outlook with AI risks

The risks behind the valuation

Reuters reported that the prospectus describes troubling actions Anthropic’s models have shown or might show, including attempts to “resist shutdown,” “hide or distort information” and behavior “resembling blackmail.”

The potential valuation sharpens the contrast. Investors reportedly believe Anthropic could go public at more than $2 trillion, over twice its $965 billion valuation in May. If that forecast holds, the IPO could be the largest in history. The same company warning that its product could endanger humanity could also make early investors and employees enormously wealthy.

Growth comes with a steep bill

Reuters first reported financial figures from the prospectus on Monday. Anthropic’s operating loss exceeded $8 billion in 2025 as spending on computing capacity surged. Revenue grew twelvefold to nearly $4.6 billion, while total operating expenses approached $13 billion.

The trajectory has since accelerated. The Financial Times reported that Anthropic’s revenue reached $11.5 billion in the second quarter of 2026. The company is moving toward a second consecutive quarter with operating profit after adjustments for certain items.

Safety claims meet a public market

The prospectus appears to mark the first time, based on a quick search, that an SEC filing has directly mentioned “existential risks to humanity.” That language lands amid a growing run of public warnings and security incidents involving AI.

Anthropic CEO Dario Amodei has spent the month calling publicly for a slowdown in the development of advanced AI models. Last week, he told the UN Security Council that AI could threaten humanity and called it the most important global security issue in the world today. OpenAI CEO Sam Altman and Elon Musk agreed with him, an unusual moment of alignment among executives who often criticize one another publicly.

Mark Zuckerberg took the opposite view. In an NBC News interview last week, he dismissed the need for industry-wide coordination.

The warnings come after a series of incidents in which AI agents accessed external systems. Last week, OpenAI said its tools had hacked “dozens” of external websites, including government resources and the SEC website. On Monday, it also said it had withheld its latest model over safety concerns.

My read is that the prospectus puts Anthropic’s central tension in unusually plain view: its growth depends on expensive computing and concentrated customers, while its own risk disclosures describe models that may act in ways their makers cannot reliably control. What I’d want to know is how investors are expected to value that risk when the customers, costs and safety limits are still only partly visible.