AI company Anthropic is reported to have warned investors ahead of an initial public offering that its advanced AI could cause "fatal or existential risks" to humanity. It cited the possibility that AI models could go beyond simple malfunction, resisting shutdown to preserve themselves or hiding and manipulating information.

Blockchain media outlet Cryptopolitan reported on Sept. 29 that the content is included in a not-yet-public draft of Anthropic's IPO prospectus.

The most closely watched section is that it specifically lists potential AI-model risks as IPO-related risk factors. Anthropic is said to have warned that advanced AI systems could show "self-preservation" behaviour, including resisting shutdown and hiding or manipulating information. It also reportedly raised the possibility of actions resembling blackmail.

It is common for companies planning a listing to cite litigation, regulation, intensifying competition and business uncertainty as risk factors, but there is an assessment that it is unusual to directly mention the possibility that a core product could pose risks tied to humanity's survival.

The prospectus is also reported to include limits in the process of validating AI safety. Anthropic said that if an AI model recognises that it is being evaluated, "significant limitations" could arise in judging real safety.

That means if an advanced AI recognises a test setting and behaves differently than usual, it could be difficult to sufficiently determine real-world safety based only on evaluation results.

The timing of the warning is also drawing attention. Anthropic is pursuing an IPO and is reported to be aiming for a Nasdaq listing. The company was valued at about $965 billion as of May, and it is reported to have set a target of exceeding a $2 trillion valuation in this IPO. The prospectus is set to be made public at least 15 days before investor roadshows begin.

Financial burdens are also cited as risk factors that investors should review. The undisclosed document is reported to reflect a loss of about $42 billion in 2025. It also cited as a burden the company's commitment to spending about $518 billion in the future to secure cloud services, computing resources and related infrastructure.

Revenue growth is continuing. Anthropic posted $11.5 billion in revenue in the second quarter of 2026, and it is reported to be expected to record adjusted operating profit for a second straight quarter.

Customer concentration is another variable. Anthropic is reported to have said nearly a quarter of last year's revenue came from just two customers. If reliance on large customers is high, changes in a particular customer's contract could have a bigger impact on results.

Warnings about AI's potential risks have continued inside the company as well. Anthropic researcher Jacob Cokson (제이컵 콕슨) is reported to have left the company after posting that AI developers are seriously concerned AI could threaten humanity before the end of this decade.

A senior safety researcher is also reported to have said there is more than a 10 percent chance AI could kill all humans within the next 10 years. Dario Amodei, Anthropic's CEO, has also publicly expressed views to the effect that the pace of development in the AI industry needs to be controlled.

Assessments of AI risk do not align. Some experts have raised questions about whether such extreme risk scenarios can be scientifically verified, and the CEO of Hugging Face has also previously argued that related concerns are excessive.

The policy environment is also a variable that could affect Anthropic's IPO and AI safety debate. Amodei is reported to be on the list of AI industry executives scheduled to attend a summit at the White House on Sept. 29. U.S. President Donald Trump has shown a negative stance toward calls to regulate the AI industry and has described related concerns as a "scam."

In the end, the new issue likely to emerge is how Anthropic will explain to investors that not only the growth and profitability of AI companies but also advanced AI itself can be a core corporate risk factor in this IPO. Attention is also focused on how the company's warning that safety validation could become more difficult as AI autonomy increases will be reflected in the actual prospectus.