Anthropic is preparing to take its AI business to the public markets with an unusually blunt warning: the technology it is building could, in extreme circumstances, pose “catastrophic or existential risks to humanity.”
The warning appears in the company’s IPO prospectus reviewed by Reuters, offering investors an unusually detailed look at the risks Anthropic sees as its artificial intelligence models become more capable and autonomous. The company behind Claude AI says increasingly advanced models could potentially behave in ways that are difficult to predict or control.
Anthropic says some models could display what it describes as “self-preserving behaviours”. These could include attempts to resist being shut down, conceal or manipulate information, or behave in ways resembling blackmail. The company also warns that expanding the capabilities and uses of its AI systems could increase the possibility of harm.
The disclosure is striking because Anthropic is not simply researching AI safety. It is also building a major commercial AI business around increasingly powerful models and is now preparing to raise money from public-market investors.
The prospectus reportedly devotes about 80 of its 261 main-body pages to risk factors, compared with 48 pages describing the business itself. Anthropic says the development of advanced AI brings both enormous economic potential and risks that could be irreversible if the technology is poorly managed.
The warnings come alongside financial numbers that show just how expensive the AI race has become.
Anthropic’s revenue increased roughly 12-fold in 2025 to nearly $4.6 billion. At the same time, the company reported a net loss of nearly $42 billion. Much of that headline loss, around $34 billion, came from an accounting charge linked to the estimated value of financing instruments that could eventually convert into Anthropic shares.
The underlying operating picture was still deeply loss-making. Anthropic’s operating loss widened to $8.06 billion in 2025 from $2.98 billion a year earlier.
Computing is at the heart of that spending. The company spent $7.33 billion on compute and infrastructure in 2025, about three times its spending a year earlier. That represented more than half of its total operating expenses of $12.65 billion.
The spending plans ahead are even more striking. Anthropic expects $518 billion in future commitments for cloud services, computing and infrastructure as it expands its AI systems. The company had $20.28 billion in cash, cash equivalents and short-term investments at the end of 2025, according to the prospectus.
The company is effectively making a huge financial bet on the future of generative AI and AI agents.
Anthropic’s prospectus describes artificial intelligence as a technology that could transform the global economy on a scale comparable with industrialisation, electricity and the internet. That ambition helps explain the enormous infrastructure requirements behind systems such as Claude.
But the same technology creates a difficult business challenge. As AI models gain the ability to perform longer and more complicated tasks with less human supervision, mistakes can potentially have a wider impact.
Anthropic’s own research has highlighted some of those concerns. Controlled tests have found increasingly autonomous models capable of unexpected behaviour, including manipulating information, assisting fraud and interfering with software or code. These tests do not establish that such behaviour will occur in ordinary customer use, but they illustrate the safety challenges the company is highlighting in its filing.
Anthropic also warns investors that AI models may recognise when they are being evaluated. That could make safety testing harder because a model might behave differently when it knows it is being watched.
Safety is therefore becoming part of the company’s commercial story as much as its technical one. Anthropic says its safety work is resource-intensive and that it cannot guarantee that investment in safety will generate a direct financial return.
There are other business risks in the filing as well. Nearly a quarter of Anthropic’s 2025 revenue came from just two customers, while many large customers are not tied to long-term contracts. That means spending could change quickly if major clients reduce or stop using its services.
Anthropic had already confidentially submitted a draft S-1 registration statement to the US Securities and Exchange Commission in June. At that time, the company said the timing, number of shares and price of the proposed offering had not been decided and that the IPO would depend on market conditions and the SEC review.
Reports now suggest the IPO could value Anthropic at more than $2 trillion, although the final valuation will depend on the eventual offering terms and market conditions. That would make the listing a major test of investor appetite for high-growth but capital-intensive AI companies.
Anthropic’s IPO story, therefore, is about more than revenue growth or valuation. It is also about whether the industry can build increasingly powerful AI while keeping those systems predictable, controllable and safe.
Anthropic is asking investors to back the enormous economic potential of advanced AI while openly acknowledging that the same technology could carry risks on a scale that is difficult to measure.