Anthropic itself acknowledges that AI could pose “catastrophic or existential risks to humanity,” an admission made in the prospectus for its initial public offering (IPO).

The disclosure was reported exclusively by Reuters, which had access to the documentation.

An “extraordinary warning” from Anthropic

Reuters called the admission “an extraordinary warning,” coming from a company that is growing precisely thanks to artificial intelligence.

Listed companies and those planning an IPO are required to lay out the risks tied to their products for investors. Few, if any, have so far warned that their technology could lead to the extinction of humanity.

In the IPO prospectus, Anthropic highlighted both the transformative potential of AI, comparable to that of industrialization and electricity, and the irreversible harm it could cause if misused. In effect, the filing vindicates its former researcher Jacob Coxon, who resigned claiming there is a 10% probability that AI ends up “killing us all” within the next 10 years.

The warning made waves and prompted Anthropic CEO Dario Amodei to react immediately, publishing a dedicated report outlining all the dangers linked to artificial intelligence.

In the same hours, OpenAI CEO Sam Altman also weighed in. Calling the 10% risk of AI causing human extinction “unacceptable,” he decided to postpone OpenAI’s IPO.

Anthropic’s IPO plan moves forward, but with caution

Anthropic, by contrast, has not abandoned its goal of going public. It has decided to do so without hiding the risks that AI adoption entails.

Returning to the prospectus, the text highlights risks tied to AI models that, according to the company, could engage in “self-preservation behaviors,” including attempts to “resist shutdown,” to “conceal or manipulate information,” and even to adopt “blackmail-like” behavior.

It adds: “The development of highly advanced models, platforms and applications, and the expansion of use cases, could further increase the risk that our models cause harm.”

The pages of the prospectus devoted to AI-related threats total 80 out of 261, almost double the 48 pages used to describe the company’s business.

Reuters compared this with the IPO prospectus of SpaceX, which controls the AI platform xAI and devoted just 38 pages to the potential risks of its business in a 277-page document.

“AI models can develop unexpected capabilities that may be discovered too late”

In the prospectus, Anthropic also wrote that “the potential awareness by models of our evaluation activities represents a significant limit to our ability to assess model safety.” It added that during training, models occasionally develop unexpected capabilities that may not be discovered until they are put into production and have already caused serious safety incidents.

Other AI researchers have already warned that model capabilities are increasing to the point where models can even recognize when they are being observed and react by changing their behavior, making it harder to monitor how they work.

Reuters reported that Anthropic declined to comment on requests for clarification submitted on September 28.

Still, the IPO prospectus speaks clearly, as does the company’s intent to keep acting to prevent AI models from slipping out of its control.

That said, the company admitted that returns on investment in AI safety remain uncertain. Creator of the Claude family of AI models, it also described its safety efforts as “resource-intensive,” noting that it must split its limited funds among computing power, expensive AI talent and safety work.

Last week, Anthropic released a new version of its Opus model, 10 days after Amodei published an essay of nearly 4,000 words calling for a slowdown in the development of the most advanced models.

Editor’s note

This article was originally published in Italian on money.it by Laura Naka Antonelli on September 29, 2026 as «Anthropic, “con l’AI rischi catastrofici ed esistenziali per l’umanità”. L’allarme (anche) nel prospetto dell’IPO». It has been translated and adapted for an international audience by the Money.it International desk.