Nobel Winners, Researchers Accuse Anthropic of Abandoning Safety Commitments

Anthropic began trading on the Nasdaq on November 19 under the ticker ANT after an $81.3 billion IPO that valued the company at roughly $1.5 trillion. On the eve of the listing, a coalition of more than 90 researchers, executives and Nobel laureates published the “MisAnthropic Statement” accusing the company of abandoning its safety commitments while racing to build recursively self-improving AI.

The statement, released November 18 by the Center for AI Safety, was signed by economics Nobel laureate Oliver Hart, physics laureate Sheldon Glashow, Turing Award winner Judea Pearl, Fields medalists Peter Scholze, James Maynard and David Mumford, Harvard psychologist Steven Pinker, legal scholars Lawrence Lessig and Laurence Tribe, and technology founders Thomas Wolf of Hugging Face and Igor Babuschkin of xAI.

Signatories span opposing camps on AI risk — from Max Tegmark and Tristan Harris, who warn of existential danger, to Guillaume Verdon and Pinker, who oppose restrictions they say would impede progress.

Letter Levels Three Charges at Anthropic

The letter makes three core charges. First, Anthropic is competing to build recursively self-improving AI despite warning that the technology could concentrate power and escape human control. Second, the company uses its safety reputation to win over investors, employees and the public. Third, Anthropic has weakened its own safety pledges and lobbied for rules that serve its interests. The signers conclude: “Anthropic is not a safe AI company”.

Safety Policy Revised Before IPO

Anthropic’s Responsible Scaling Policy, introduced in 2023, committed the company to extra oversight and security provisions once it could “compress two years of . . .” “AI progress into a single year” based on its 2018–2024 development pace. Anthropic activated part of its responsible scaling policy after its own internal estimates indicated it could compress two years of AI progress into a single year.

Protest messages against Anthropic's IPO in front of the company's headquarters in San Francisco, California. Reuters
Photo: The National

Critics point to a February revision that removed a unilateral pause commitment — the pledge to halt development independently if safety measures proved insufficient. The rewrite introduced Frontier Safety Roadmaps and recurring Risk Reports but retained only conditional commitments to delay releases when Anthropic has clear evidence it is ahead and has yet to contain a catastrophic risk.

Recursive Improvement Pitched as Competitive Edge

In its SEC filing, Anthropic highlighted a “virtuous cycle” where each Claude version helps build the next, calling it a key competitive advantage. The filing also said that internal use of AI systems “has already been accelerating our research and development” and that “we may experience a further increase in the rate of progress in AI capabilities”.

The AI Futures Project called the acceleration implied in METR’s report very unexpected.

The statement accuses Anthropic of lobbying to weaken proposed regulations in California (SB 1047), New York (RAISE Act) and at the federal level (AEGIS Act), while also advocating privately for federal preemption of state AI laws. The company did not respond to requests for comment on the MisAnthropic Statement.

Leaders Frame Safety as Strategic Advantage

Jasmine Sun, a New York Times opinion columnist, wrote that Anthropic insiders believe they must develop powerful AI first so worse actors will not, Finance Yahoo reported from a podcast discussion of her piece.

Anthropic
Photo: The Hindu

Speakers on the podcast said the company argues the most insane models cannot go out uncontrolled and must work in humanity’s interest, not the other way around. The discussion also noted Anthropic’s public benefit corporation structure and a planned donation fund, though the structure is revocable.

Nobel economics laureate Oliver Hart joined the MisAnthropic Statement as a signatory on November 18, adding his authority on contract theory and governance to the letter’s charge that Anthropic weakened its safety pledges while lobbying against regulation.

Hart, the Lewis P. and Linda L. Geyser University Professor at Harvard University since 1993, received the Nobel Memorial Prize in Economic Sciences in 2016 for work on incomplete contracts and ownership structure — research that analyzes how parties govern relationships when they cannot specify every contingency, a framework directly relevant to oversight of recursively self-improving AI systems.

IPO Prices Above Range, Trading Begins

Anthropic priced 980 million shares at $83, above the previously indicated $70–78 range. Trading began November 19.