Anthropic warns of 'catastrophic' AI risk ahead of $2 trillion IPO

What's new

Anthropic, the company behind the chatbot Claude, is heading for the stock market with one of the largest share offerings ever. In the official IPO prospectus (the document investors must read before they can buy shares), Anthropic itself writes that its artificial intelligence could pose "catastrophic or existential risks to humanity." The company is aiming for a valuation of more than $2 trillion, more than twice the $965 billion Anthropic was valued at as recently as May 2026.

What's clever

What is unusual is that Anthropic itself highlights the dangers of its own product. Of the prospectus's 261 pages, about 80, almost a third, are about risks. Here, the company writes that its AI models can exhibit "self-preserving behavior," including attempts to resist being shut down, hiding or manipulating information, and behavior resembling blackmail. This is not just a legal disclaimer in fine print; it is a direct admission that the technology can behave unpredictably.

Cheaper or better?

This is not about the price of a product, but about the business itself. Anthropic had $4.6 billion in revenue in 2025, a twelvefold increase from the year before, but at the same time had an operating loss of more than $8 billion. On the books, the total loss came to a full $42 billion, although around $34 billion of that was due to a technical revaluation of the value of future financing instruments and not ordinary operations. The company has also committed to spending $518 billion on cloud computing and data centers in the coming years, money it has not yet earned. Rival OpenAI is also deeply unprofitable, but Anthropic stands out by warning so openly about risks in its own prospectus.

What it's good at

The IPO also lays bare how dependent Anthropic is on a few large customers: almost a quarter of last year's revenue came from just two customers, who are not bound by long-term contracts. For ordinary Danes and businesses, the most important point is that even the companies that earn the most from AI now openly acknowledge that the technology could run amok, at the same time as tools like Claude are becoming more widespread in workplaces. According to the sources, the IPO may take place after the US midterm elections in November 2026.