Anthropic Mathematician Says Claude Helped Prove Riemann Hypothesis
In this article
Levent Alpöge, a mathematician at Anthropic, said on Friday, November 20 that he and the company’s Claude model had discovered a proof of the Riemann hypothesis, a problem posed by Bernhard Riemann in 1859 with a $1 million Millennium Prize attached. The announcement came one day after Anthropic began trading on Nasdaq, according to Decrypt.
Timing scrutiny after the Nasdaq debut
Alpöge acknowledged the awkward sequencing himself, writing that “the precise timing is horrible.” He rejected a cynical reading of the release date but did not say directly that the IPO caused the delay, stating only that there was a constraint to respect after he spoke with company leadership.
The result itself is the wider claim. Anthropic had previously reported progress on the zeta function without asserting a solution to the full hypothesis; Alpöge’s note goes considerably further, extending the argument to other mathematical functions through the Generalized Riemann Hypothesis.
Lean can verify that formal deductions obey its rules, but that is a narrower check than mathematical acceptance. Mathematicians must reproduce the archive, confirm that the formalized theorem matches the original conjecture, and examine the result before treating it as settled.
The shadow of September’s AI math dispute
Alpöge is not a neutral figure in the debate over machine-assisted mathematics. He featured in September’s credit dispute after OpenAI claimed it had solved the Navier-Stokes Millennium Prize problem while he and NYU mathematician Tristan Buckmaster were pursuing related fluid-equation work.
OpenAI followed in October with 722 manuscripts generated by an unreleased model, grouped into 372 families of results. Only 162 included Lean-checked main results, and mathematicians asked for stronger evidence rather than accepting the volume of claims.
Some researchers now worry about a “mathocalypse,” in which AI solves prized problems before people can learn from working on them. Toronto mathematician Daniel Litt warned that without adaptation there would be “no more math in 50 years,” while Buckmaster argued at an MIT panel this week that mathematics still depends on human understanding. If Alpöge’s claim holds, it will be judged against that backdrop as much as against the 167 years of work behind the problem.
Why it matters
A verified proof would settle a question that underpins the study of prime numbers, and it would also reset expectations for what AI models can contribute to pure research. Until then, the episode tests whether AI companies can release bold mathematical claims in a way the research community considers credible, and whether publication timing near a corporate milestone changes how those claims are read. BitcoinWorld has previously covered how AI agents perform on technical benchmarks, including a comparison of CoinStats, Gemini, Claude and ChatGPT on crypto research, which shows how quickly benchmark claims in this space invite scrutiny.
What to watch
The next signals are independent reproduction of the 14-part Lean archive, any submission to a qualifying journal, and public reaction from mathematicians who worked on the problem. The Clay Mathematics Institute’s two-year scrutiny window means no prize decision can arrive quickly, whatever the archive shows.
Source: Decrypt




