The 167-year-old Riemann Hypothesis, one of mathematics' most enduring unsolved problems, may be on the verge of its first major breakthrough — but whether an AI-generated proof can withstand scrutiny from the global mathematics community remains the biggest open question.
On November 20, Anthropic mathematician Levent Alpöge issued a statement claiming that he and the AI model Claude had found a method to prove both the Riemann Hypothesis and the Generalized Riemann Hypothesis (GRH), releasing more than 17 million lines of Lean formalization code alongside the announcement. Because the declaration came just one day after Anthropic's stock market debut, the mathematical breakthrough also sparked discussion about the timing of the release and possible commercial motivations.
Alpöge wrote the proof in the Lean programming language and published it across 14 compressed files. Meanwhile, the text-based proof he generated with Claude exceeded 250,000 pages, but was not made public due to its overwhelming length and readability challenges.
At present, no mathematician has formally confirmed that the proof holds. Independent verification is still needed to determine whether the theorem ultimately proved by the Lean code genuinely corresponds to the Riemann Hypothesis itself, and whether any logical gaps exist in the overall proof process.
How the proof was achieved: Claude's critical breakthrough at the finish line
In his statement, Alpöge described the core process behind the collaboration: he pointed Claude toward a key research direction, and Claude subsequently proposed what he considered a highly groundbreaking critical step — a "theta elevation" chain whose length is approximately equal to the square root of the sum of the defined prime numbers. Each elevation corresponds to the "multiplicity" that classical function field arguments rely on, and this multiplicity is extremely difficult to identify over the integer field.
"Near the finish line, Claude far surpassed me. No — Claude was practically flying, dancing, and shining near the endpoint," Alpöge wrote.
He also attached what he said was a research timeline compiled by Claude: the work began on October 6 with the "Stepanov determinant," and by November 16 only one entry remained — "GRH."
MIT mathematician Andrew Sutherland, after independently compiling the relevant Lean code, said: "This is indeed a historic moment. The world has already changed... In my view, this is a victory for humanity, a day worth celebrating."
Reaction from the mathematics community: cautious optimism, awaiting independent verification
Multiple leading mathematicians and AI researchers have expressed interest in the development, but remain cautious about whether the proof truly holds.
Terence Tao, winner of the 2006 Fields Medal, wrote on his personal blog that preliminary analysis suggests the proof appears to combine modern research results from the Langlands program and arithmetic geometry, and that verification work is expected to take some time.
Mathematician Elliot Glazer's initial inspection found that the mathematical proposition formalized in the Lean files is indeed the Generalized Riemann Hypothesis, and he estimated the probability that Lean itself contains a flaw at "well below 1%." Alex Kontorovich, head of the mathematics department at Rutgers University, also said that as more details emerge, he has growing confidence in the likelihood that the proof will succeed.
A mathematics and AI researcher at Princeton University emphasized that human involvement remains the best way to use large language models to solve mathematical problems. He also noted that in this research, Alpöge deserves recognition on equal footing with Claude.
The 250,000-page proof: the tension between AI breakthrough and human verification
The enormous length of the proof has become another focal point for the mathematics community.
University of Manchester mathematician Thomas Bloom said on X that he could not decide what shocked him more: that AI claimed to have proved the Riemann Hypothesis, or that AI claimed to have generated a logically coherent proof exceeding 250,000 pages when converted into human-readable form.
Pierre Richemond, head of multimodal business at Cohere and a former Google DeepMind researcher, argued that the Riemann Hypothesis may already have been solved, but that Lean verification is still pending. This development raises further questions: AI-generated research output now carries serious mathematical value, yet it has not been adequately verified by the human mathematics community.
Timing controversy: announced the day after IPO, Alpöge responds directly
The timing of the announcement has also drawn outside attention. Anthropic went public on November 19, and Alpöge revealed this major mathematical advance the following day, inevitably prompting speculation about commercial publicity motives.
Alpöge addressed the controversy directly in his statement, acknowledging that "the precise timing is terrible." He said the key work was completed late at night on November 15 local time into the early hours of November 16, after which he asked Anthropic CEO Dario Amodei whether he could make it public. "Ultimately we figured out that there were indeed restrictions that needed to be observed," he said.
He denied negative speculation about the release timing, calling such "obvious cynical views" mistaken, and said he "simply could not wait any longer," adding that he was willing to take full responsibility.
It is worth noting that Anthropic had already made incremental progress in AI mathematical research. In August of this year, the company published Claude's research findings on the distribution of zeros of the Riemann zeta function, raising the lower bound on the proportion of zeros on the critical line from 41.6% to 67.2%. At the time, however, Anthropic explicitly stated that it did not expect the techniques used by Claude to ultimately prove the Riemann Hypothesis.
Prize and recognition: still years away from the million-dollar award
Even if the proof is ultimately confirmed, Alpöge remains a considerable distance from the $1 million prize established by the Clay Mathematics Institute.
Under the institute's rules, a solution to a Millennium Prize Problem must undergo formal publication, at least two years of public scrutiny, and gain general acceptance from the international mathematics community before a prize can be claimed.
Alpöge himself is optimistic about the form the proof may take once it is eventually accepted. He believes the proof will likely be substantially simplified, and that this could happen soon. He described the development as "a dawn of something greater in the world of science."
Capital markets have yet to form a clear judgment on the news. Odds on prediction platforms still reflect divided market sentiment. Meanwhile, an old post by OpenAI CEO Sam Altman on X from 2022, in which he half-jokingly asked "when will the Riemann Hypothesis be solved," was dug up by netizens, with one commenter quipping: "Anthropic beat you to it, buddy."