In October 2026, Fields Medalist Terence Tao told Numberphile that solving the Riemann Hypothesis would require something beyond the AI systems available at the time. He suggested that humans and AI might get there around 2050.
Then November happened.
On November 20, mathematician Levent Alpöge announced a proof of the Riemann Hypothesis developed with Claude. The next day, the Lean and Mathlib communities said they had independently checked the formal proof and certified its correctness.
Five weeks. That's how much time passed between the interview and one of the most extraordinary AI-assisted mathematics announcements we've seen.
WHAT ACTUALLY HAPPENED?
• October 15: Terence Tao discusses the limits of AI mathematics with Numberphile.
• November 20: Levent Alpöge releases a roughly 17-million-line proof in Lean, developed with heavy assistance from Claude.
• November 21: The Lean and Mathlib communities announce that the statement and proof passed formal verification.
THE IMPORTANT DETAIL: THIS WASN'T AN AI WORKING ALONE
Alpöge is a mathematician. He directed the research, worked with Claude, and released the resulting formalization.
In his announcement, he described Claude as doing much of the work near the finish line. That is a remarkable account of human-AI collaboration, but it is not evidence of a fully autonomous AI discovering and independently publishing the proof.
WHAT DID LEAN VERIFY?
Lean is a proof assistant. Instead of judging whether a mathematical argument sounds convincing, its kernel checks that each formal step follows from accepted rules.
The Lean and Mathlib communities reported that:
• The formal statement matches the actual Riemann Hypothesis.
• The 17-million-line project compiles.
• Independent comparison tools confirmed that the code proves the intended theorem.
• Lean's kernel and several other proof checkers independently accepted the exported proof.
That is much stronger evidence than an AI-generated paper claiming to solve a famous problem.
DO WE UNDERSTAND THE PROOF?
Not fully yet.
The code is enormous, and it wasn't accompanied by a conventional, readable mathematical exposition. Researchers still need to identify the central ideas, simplify the argument, and explain why the method works.
Tao himself highlighted that distinction on November 20: formally establishing a result and digesting the mathematics behind it are different jobs.
And the Clay Mathematics Institute has its own recognition process for Millennium Prize Problems. Formal verification does not instantly mean the million-dollar prize has been awarded.
WHY THE RIEMANN HYPOTHESIS MATTERS
Proposed in 1859, the hypothesis concerns the nontrivial zeros of the Riemann zeta function and their relationship to the distribution of prime numbers.
It's one of the most famous open problems in mathematics, and one of the Clay Millennium Prize Problems.
The extraordinary part isn't that Tao was foolish to be cautious. It's that the capabilities of AI-assisted mathematics are moving fast enough to make even reasonable expert estimates age very quickly.
A month ago, 2050 sounded plausible.
Today, the question is no longer just whether AI can help produce a formally verified proof.
It's whether humans can understand the mathematics it helps discover.
SOURCES
Lean and Mathlib verification statement:
https://leanprover-community.github.io/blog/posts/riemann-hypothesis/
Levent Alpöge's announcement:
https://alpo.ge/remark.pdf
Terence Tao's November 20 commentary:
https://terrytao.wordpress.com/
Original Numberphile interview:
https://www.numberphile.com/podcast/terence-tao-ai
Clay Mathematics Institute:
https://www.claymath.org/millennium/Riemann-Hypothesis/
Are we entering an era where AI can help prove theorems faster than mathematicians can explain them?
#RiemannHypothesis #ClaudeAI #TerenceTao
YouTube / ABVX
