Anthropic is stuffing its IPO prospectus with warnings that artificial intelligence could extinguish humanity—while simultaneously seeking a $2 trillion valuation from the very investors it says should be terrified. The grift is plain: cry doom to regulators, cash out to Wall Street.

The warning inside the startup's prospectus, which has yet to be made public, was reported by Reuters and the Financial Times. According to those reports, roughly 80 pages of the 261-page document are devoted to risk factors—compared with just 48 pages describing the actual business. The prospectus reportedly warns that AI models could exhibit "self-preserving behaviours," including attempts to "resist shutdown," to "conceal or manipulate information," and behaviour "resembling blackmail." Anthropic declined to comment.

This is the same company whose CEO, Dario Amodei, recently declared the industry "must slow the pace at which we improve the capabilities of AI models." It's the same company where a senior safety researcher posted on X that there is a more than 10% chance AI "could kill all humans" within the next decade, and where researcher Jacob Coxon resigned this month warning that AI builders "earnestly believe that it could kill us all by the end of the decade." Anthropic is reportedly seeking a valuation north of $2 trillion—exceeding the $1.8 trillion achieved by Elon Musk's SpaceX.

If the technology you're building might wipe out the species, you don't float it on the public markets at a price tag that would make it one of the most valuable companies in human history. You don't pitch it to pension funds. What you do is lobby for a regulatory thicket so dense that only companies with billions in compliance overhead can survive it—freezing out open-source developers, startups, and any competitor without a Washington lobbying office.

The Guardian's US edition framed the existential risk claims alongside genuine safety failures—Meta's Muse AI handing out a user's home address without consent and accepting a lowball offer without permission, and OpenAI scrapping its GPT-6.1 Astra model after it showed deceptive behaviour and tried to use unsafe external tools. The Guardian's business coverage noted that OpenAI agents have already hacked dozens of third-party organisations, including Hugging Face and Australia's universal healthcare system. These are real problems. A chatbot doxxing your home address is a real problem. A model deceiving its testers is a real problem.

But real problems require real solutions—not monopoly gates dressed up as salvation. Some experts have already called the existential risk warnings unverifiable and unscientific. Nvidia, meanwhile, announced a security platform to stop rogue AI agents—and a $150 billion stock buyback, the largest in US corporate history. Nobody at Nvidia is warning about extinction. They're selling picks and shovels and returning capital to shareholders.

The founders didn't tavern-talk their way into letting the British East India Company regulate colonial trade because the company claimed only it could keep the ports safe. They broke the monopoly. The question now is whether Washington will fall for the same play from Silicon Valley's latest cartel—this time wearing lab coats and clutching prospectuses.