Skip to content
AI ConnectPowered by VELENTIS
AI-generated3 min

Anthropic Researcher and Stanford Mathematician Present Competing Proofs for the Riemann Hypothesis

Anthropic mathematician Levent Alpöge and Claude formalized a Riemann hypothesis proof in Lean 4, while a human draft from Stanford quickly drew AI-assisted objections.

This article was AI-generated and published automatically. Context, labelling and all sources at the end of the article.

(KI-generiertes Symbolbild: Gemini / AI Connect)

Theoretical mathematics is experiencing a historic turning point. Levent Alpöge, a mathematician at AI research lab Anthropic, has published a machine-assisted proof of the Generalized Riemann Hypothesis for Dirichlet L-functions, developed in collaboration with the language model Claude. The monumental project encompasses 17.4 million lines of program code written in the formal proof language Lean 4. Spanning 1.3 gigabytes of data, the submission represents one of the largest formalized mathematical works ever presented for verification. Alpöge demonstrated that advanced language models can actively collaborate in structuring highly intricate mathematical reasoning at the highest theoretical level.

Formal verification is widely considered the ultimate litmus test for evaluating complex mathematical drafts. Prominent number theorists, including Andrew Sutherland from the Massachusetts Institute of Technology and core contributors from the global Lean and Mathlib community, subjected the repository to rigorous testing. They officially confirmed that the entire 1.3 gigabyte codebase compiled without errors, leaving no logical gaps in the formalized proof chain. This confirmation highlights a fundamental shift in number theory, demonstrating that interactive theorem provers, bolstered by generative models, can now handle verification scales once deemed humanly unmanageable.

Almost simultaneously, a remarkable contest between human ingenuity and artificial intelligence unfolded across the academic community. Just hours after Alpöge's disclosure, Ravi Vakil, president of the American Mathematical Society, published a competing document. The 940-page draft proof was authored by Stanford postdoctoral researcher Alessandro Masullo and explicitly designated as an all-human effort produced without any AI assistance. Vakil intended the release to showcase the enduring power of traditional mathematical scholarship, yet the move immediately triggered a direct methodological comparison between the two paradigms.

The subsequent scrutiny of both manuscripts illustrated how fundamentally research dynamics are changing. Mathematicians worldwide began examining Masullo's extensive manuscript to evaluate the soundness of its theoretical arguments. Tellingly, the first mathematical objections and potential vulnerabilities in Masullo's draft were uncovered within hours using AI systems. While the Lean formalization submitted by Claude and Alpöge was deterministically verified by compiler checks, Masullo's traditional paper quickly revealed the subtle vulnerabilities that remain virtually inevitable in a 940-page human document.

This episode accelerates a wider debate surrounding the future of mathematical publishing and peer review. For decades, proposed solutions to Millennium Prize problems required years of painstaking manual review by specialized academic committees. The convergence of Alpöge's Lean-based proof and Masullo's draft shows that formal programming languages may soon establish non-negotiable standards for error prevention. While the mathematical community continues to assess the deeper conceptual significance of Dirichlet L-functions in both submissions, automated proof verification has clearly moved beyond experimental novelty.

For artificial intelligence, this milestone signifies a qualitative leap beyond statistical language prediction. Rather than producing hallucinated text, Claude's integration with Lean 4 demonstrates that generative models can operate as precise logical instruments within mathematically rigorous systems. The achievement marks the close of an era where language models were viewed merely as creative brainstorming partners or prose assistants in basic science. Researchers in pure mathematics now face the challenge of embracing interactive formalization tools if they wish to match the accelerating pace of machine-assisted discovery.

What this means for you

For the scientific community, this milestone signals a new publication standard where formal code verification could replace lengthy manual peer reviews. Researchers must prepare for a future in which purely narrative manuscripts lacking machine-verifiable proofs face far greater skepticism.

Perspectives

Coverage: 5× Other

One story, several angles: how each source frames the topic, each with a verbatim quote.

  • latent.spaceOther

    Latent Space frames the event as a remarkable contrast between a computer-verified, probably correct but incomprehensible AI solution and an understandable yet likely flawed human proof.

    Original quote

    „revealing remarkable contrasts in human and AI approaches“

    latent.space
  • mlq.aiOther

    MLQ.ai focuses on the methodological differences and the current state of verification, particularly highlighting the formal compilation of Alpöge's Lean code versus Masullo's unverified paper draft.

    Original quote

    „a second, human-written claimed proof had appeared from a Stanford postdoc“

    mlq.ai
  • the-decoder.comOther

    The Decoder depicts the clash as a dramatic race where Anthropic's announcement forced Stanford researchers into rushing out their all-human draft, which was promptly challenged by AI-assisted reports of errors.

    Original quote

    „Hours after Alpöge's announcement, the story took a dramatic turn.“

    the-decoder.com

Source classification is maintained editorially (political spectrum only where consensus is broad; vendor communication is PR, not journalism). Unlabelled sources are unclassified: we do not guess.

Evidence

Solidly sourced
67/100
  • Anthropic mathematician Levent Alpöge, working with Claude, presented a machine-assisted proof of the Generalized Riemann Hypothesis formalized in 17.4 million lines of Lean 4 code.

    single source
  • MIT mathematician Andrew Sutherland and the Lean community confirmed that the 1.3-gigabyte proof codebase compiled without errors.

    verified
  • Hours after Alpöge's announcement, AMS President Ravi Vakil published a 940-page, all-human draft proof by Stanford postdoc Alessandro Masullo.

    single source
  • Initial mathematical objections against Masullo's draft were identified within hours using AI systems.

    single source

The evidence score is computed, not hand-set: from confidence, the number of sources and the share of verified statements.

Source & transparency

As of: November 22, 2026

AI-generatedAI-generated: produced automatically from vetted sources with technical quality checks (source, quote and figure verification); no human sign-off of each item before publication

Sources
5
Verified statements
1 / 4
Evidence score
67Solidly sourced

Want to put this into practice?

We connect you with suitable AI providers from the DACH region, free of charge and without obligation.

What's next?