SIAgentNews.com
← All news

Anthropic's 250,000-page AI math proof remains unverified

Anthropic claims to have solved the Riemann hypothesis using AI-generated code, but the sheer scale of the proof makes human verification nearly impossible.

· First reported by Emerald Book

250,000pages of AI-generated math text

The short version

  • The proof consists of 17 million lines of Lean code and over 250,000 pages of AI-generated text.
  • Anthropic's mathematician Levent Alpöge described the massive output as a 'massive unreadable travesty of slop.'
  • The Clay Mathematics Institute requires a two-year period of public scrutiny before awarding the $1 million prize.
  • Previous AI-generated math proofs, such as those for the Navier-Stokes problem, were found to contain subtle translation errors.

Why it matters

This story highlights a growing 'verification debt' where AI can produce complex results faster than humans can audit them. It raises questions about whether some mathematical truths might become inaccessible to human understanding as they scale.

The Riemann Hypothesis claim

Anthropic recently announced that it has solved the Riemann hypothesis, a famous mathematical conjecture regarding prime numbers. For 167 years, this problem has been considered one of the most difficult challenges in the field. The company's mathematician, Levent Alpöge, presented the solution on November 20, 2026, using a combination of human oversight and Claude AI.

The proposed solution is not a standard academic paper that a human could easily read. Instead, it is a massive collection of 17 million lines of Lean code. Lean is a programming language specifically designed to mechanically verify logical arguments in mathematics.

While the computer confirms that the logic compiles without errors, the mathematical community remains skeptical. The sheer volume of the output makes it impossible for independent experts to quickly confirm if the proof is correct. As of now, the Clay Mathematics Institute still lists the problem as officially unsolved.

The verification gap

A major hurdle in this announcement is the gap between computer verification and mathematical truth. Just because a computer says the logic is sound does not mean the AI solved the correct problem. Mathematicians must ensure the AI did not subtly change the definitions or include hidden assumptions during the process.

This issue was seen in September 2026 when OpenAI announced an AI-generated solution to the Navier-Stokes Millennium Prize problem. Experts later discovered that the AI had mistranslated the human-readable paper into the verification code. The computer checked a slightly different version of the theorem than the one the company claimed to prove.

Alpöge warned that the Anthropic output is a 'massive unreadable travesty of slop.' This suggests that the AI may have produced a brute-force result that lacks human-understandable clarity. The burden of proof now shifts to a small group of specialists who must hunt for flaws in millions of lines of code.

The scale of the output

The sheer size of the Anthropic proof creates a significant bottleneck for traditional peer review. The output spans more than 250,000 pages, which is roughly equivalent to 500 full-length novels. No human can read, check, or fully understand a document of this magnitude in a reasonable timeframe.

To verify this, mathematicians will have to use a 'trust the kernel, audit the edges' approach. They will rely on the computer to check the bulk of the logic while manually inspecting the top-level definitions. They must also ensure the code does not use 'cheat' commands that tell the computer to skip specific steps.

This creates a situation of verification debt where AI can generate dense math in weeks. However, auditing the results still takes months or years of grueling effort from elite specialists. The community must now figure out how to compress this brute-force output into a human-sized academic paper.

A history of false alarms

The Riemann hypothesis is known to attract many false solutions every year. Major journals and preprint servers see dozens of proposed proofs that are eventually rejected. Many of these attempts fail because they rely on generic tricks rather than the actual properties of prime numbers.

History is full of examples where proposed proofs fooled experts before being debunked. In 1945, Hans Rademacher's proof fooled Carl Ludwig Siegel until a flaw was found. In 2008, Xian-Jin Li's proof was retracted within days after a fatal error surfaced during peer scrutiny.

Other notable attempts include a 124-page proof by Louis de Branges in 2004 and a five-page lecture by Sir Michael Atiyah in 2018. Because the Riemann hypothesis looks simple on the surface, it often acts as a magnet for these types of false alarms. Anthropic's claim faces the same high level of skepticism as these previous attempts.

What comes next

For the Clay Mathematics Institute to award the $1 million prize, the proof must survive two years of public scrutiny. It must be published in a recognized venue and gain broad consensus from the mathematical community. Currently, none of these conditions have been met for the Anthropic announcement.

Mathematicians will likely use AI tools to map out the 17 million lines of code as a giant dependency graph. This process will help isolate the key breakthroughs from the routine algebraic grinding. Only then can the core lemmas be translated into a normal paper that humans can actually read.

The situation leaves the community wondering if the hardest problems in the universe require proofs too large for the human brain to hold. Until the verification is complete, the Riemann hypothesis remains unsolved. Anthropic's work remains an unverified claim that may take years to fully audit.

What to watch

  • Whether independent mathematicians can successfully audit the 17 million lines of Lean code.
  • If a human-readable version of the proof can be distilled from the 250,000-page output.
  • How the mathematical community handles the 'verification debt' of AI-generated proofs in the future.

Try it

Look up the Clay Mathematics Institute's list of Millennium Prize Problems to see the official requirements for solving the Riemann hypothesis.

Sources and links

Read the original at Emerald Book

First reported by Emerald Book. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news