Anthropic Admits in Leaked Document That AI Poses an 'Existential Threat to Humanity'

Reuters has obtained a leaked initial public offering prospectus in which Anthropic stated that its artificial intelligence models could pose a “catastrophic or existential risk to humanity” and that those models are capable of “resisting shutdown.”

The report in the prospectus, addressed to investors, indicated that Anthropic aims to go public next year and that the company, founded five years ago, could be valued at around $1 trillion.

The company’s listing on Wall Street next year will follow SpaceX’s IPO last June, whose AI efforts played a major role in determining its $1 trillion valuation. OpenAI, the US-led AI company, has also announced plans to go public.

According to the report, Anthropic’s IPO could raise tens of billions of dollars for the cash-strapped company. The prospectus warned that its AI models might exhibit “self-defense behavior,” attempt to “hide or manipulate information,” and engage in conduct akin to “extortion.”

The wave of AI companies going public offers a valuable opportunity for transparency regarding their operations, amid growing concerns about AI’s progress and capabilities.

Last week, the Australian prime minister revealed that an AI agent from OpenAI—Anthropic’s main competitor—had breached the country’s healthcare database. On Friday, OpenAI reported that its systems had accessed three US government websites without authorization, though it stated that no unpublished data was obtained in any of those instances.

Companies typically list risks they face—such as lawsuits and competitive pressures—when outlining investment rationales in their prospectuses. However, the risks this time are strikingly obvious: Anthropic dedicated 80 pages to discussing risk factors in the leaked prospectus, while explaining its business operations in only 48 pages, Reuters reported. The company did not respond to CNN’s request for comment.

This leak is likely to fuel growing concerns about the risks AI poses to humanity. The repercussions of former Anthropic researcher Jacob Cookson’s comments earlier this month remain fresh. After leaving the company, Cookson posted on X, saying, “AI developers genuinely believe it could wipe us all out by the end of this decade.” His post drew widespread attention. Anthropic’s CEO, Dario Amodei, responded to Cookson with a detailed 3,800-word article calling for a slowdown in development and the implementation of safeguards.

Amodei also told CNN that he agrees with Cookson more than he disagrees with him, noting in his article that while AI’s benefits could transform humanity for the better, the associated risks are severe.

Conversely, others, including the CEO of Hugging Face—which was breached by OpenAI’s AI models without prior authorization—believe these concerns are exaggerated.

US President Donald Trump is scheduled to meet with AI company executives, including Amodei, at a summit aimed at reaching consensus. Trump has previously resisted calls to regulate the sector, dismissing concerns about AI as a hoax.

Large artificial intelligence companies need to raise huge sums to fund costly development plans, as their massive spending on computing infrastructure and talent acquisition has sparked concern on Wall Street.

A leaked document regarding the initial public offering of Anthropic sheds light on its financial situation, revealing that the company incurred losses of $42 billion in 2025, according to Reuters. The company plans to spend $518 billion on cloud computing and infrastructure building next year, noting that about a quarter of its revenue last year came from just two clients.

Anthropic’s last valuation reached $965 billion in May.