Gravely Concerned That Uncontrolled Growth In AI Poses A Threat To Society, Says Association Of Human Mathematics On Riemann Hypothesis Proof

The Association for Human Mathematics (AHM) has weighed in on this week’s claimed proof of the Riemann Hypothesis, and it is not celebrating. In a statement posted to its website, the group argues that an AI-generated Lean program is no substitute for a proof that a human can actually be held responsible for, or even understand. “We are gravely concerned that this uncontrolled growth poses a threat to society writ large,” it says.

The AHM is barely two months old, but it has already built a reputation as the most combative voice pushing back on AI’s encroachment into mathematics. The group was formed in September around the idea that mathematical progress depends on human expertise, writing, and judgment, and that handing those over to corporate AI models hollows out the discipline from the inside. Its first real fight came in October, when it accused OpenAI of showing total disregard for scientific norms after the company dumped more than 700 AI-generated manuscripts onto GitHub in a single release, calling it a demonstration of power rather than scholarship and urging mathematicians to cut ties with the lab. That statement was reposted on Terence Tao’s blog and effectively endorsed by him. The group has since grown past a thousand members, including Fields Medalist Peter Scholze, with over 600 going further and pledging to do entirely AI-free research.

This week’s target is Anthropic rather than OpenAI, after mathematician Levent Alpöge announced a Claude-assisted proof of the Riemann Hypothesis on November 20, compiled over a 41-day stretch and credited largely to the model itself. The AHM calls it “a major departure from traditional methods in mathematical research, representing a new value system put forward by AI corporations that foregrounds technical validity of logical statements over accountability, clarity, accessibility, and understanding.” Its statement goes further: “At this time, no human being can reasonably claim to be responsible for, or even to understand, such a document. Neither can any mathematician confirm with full certainty that the proof is correct.” It adds that “this corporate model of proof production seems designed to excuse humans from responsibility for scientific work” and instead “asks scientists to entrust their research to machines that AI corporations control”.

The group frames this as something bigger than one disputed proof. “Autonomous computer programs that investigate deep mathematical questions are now instruments of power in scientific research, and scientific power enables other types of power that can directly threaten human lives,” the statement reads. “This power is now concentrated in the hands of AI companies that seem increasingly untethered from the interests of human society.” It goes on to say that “milestones such as the Riemann Hypothesis are used by AI companies as little more than publicity for the superintelligence that they wish to build,” and that this is why, in the group’s view, “AI corporations urgently require democratic oversight.”

The statement also returns to a theme the AHM has hit before: that an avalanche of machine-generated results, however technically sound, can’t replace human mathematical work. Claude’s history with this specific problem goes back further than this week — an earlier, unreleased version had already pushed a related bound from 41% to 67% back in August without cracking the hypothesis itself, a result Anthropic was careful at the time not to oversell. The AHM’s objection isn’t really about whether that chain of work eventually got there; it’s about what gets lost along the way. “As a model for scientific research, the overproduction of unreadable mathematical documents pales in comparison to the richness of the mathematical dialogue that humans have conducted for millennia,” the statement says, adding that while “today’s announcement builds upon over a century of human work,” it “fails to advance mathematical understanding in the same way.”

Whether this moves the needle on how Anthropic or the wider field treats AI-assisted proofs remains to be seen. But with the AHM’s membership still climbing and a second major AI lab now on its list of grievances, the fight over who gets credit for solving mathematics’ hardest problems looks far from settled.

Posted in AI