AI Helps Refute 90-Year-Old Jacobi Conjecture in Mathematics 2026

The Jacobi conjecture, a nearly 90-year-old mathematical problem, has been refuted with the help of an AI model called Fable 5, according to a tweet by Harvard mathematician Levent Alpöge. The breakthrough highlights the growing role of artificial intelligence in solving complex mathematical challenges.

While Spain was making history by winning its second FIFA World Cup on Sunday, Alpöge, 34, asked Fable 5—an AI model from Anthropic—to do something interesting: not to prove the Jacobi conjecture, but to disprove it. And it worked.

What Is the Jacobi Conjecture?

The Jacobi conjecture posits that a certain type of mathematical function can also work in reverse. It was formulated by Ott-Heinrich Keller in 1939 and later included by Stephen Smale in his famous list of unsolved mathematical problems for the 21st century.

For decades, mathematicians have struggled to prove or disprove it. The conjecture is part of a broader set of problems in algebraic geometry and polynomial mappings, and its resolution—or in this case, refutation—marks a significant moment in the field.

How AI Helped Refute the Conjecture

Alpöge has not explained in detail how he achieved the result or what prompt he used. However, in his tweet, he indicated that part of the credit was due to his "close friend Fable," the Anthropic model.

The use of AI in mathematical research is not entirely new, but this case demonstrates the potential of large language models to assist in discovering counterexamples or refuting long-standing conjectures. It also raises questions about the role of AI in creative mathematical thinking.

Why This Matters for Mathematics

The refutation of the Jacobi conjecture is important because it closes a chapter on a problem that has puzzled mathematicians for generations. It also shows that AI can be a powerful tool in the search for counterexamples, which are often harder to find than proofs.

Moreover, this achievement could inspire other mathematicians to use AI in their own research, potentially accelerating the pace of discoveries in mathematics and related fields. The background to this story is set out in Anthropic's Claude Fable 5.1: Most Powerful AI Model Yet, But Capped. A related thread runs through Claude Fable 5.1: New Features and 25% Token Cost Drop. Earlier coverage in AI in Mathematics: Why Human Mathematicians Still Matter shows how the argument developed.

• AI as a research assistant: Fable 5 helped refute a conjecture that stood for nearly 90 years.
• Counterexample over proof: The AI was asked to disprove, not prove, the conjecture—a strategic move.
• Historical context: The conjecture was formulated in 1939 and included in Smale's list of 21st-century problems.
• Attribution: Alpöge credits the AI model as a "close friend" in his tweet.
• Implications: This case may encourage broader adoption of AI in mathematical research.

Comparing AI Models in Mathematical Research

What to Watch Next

The mathematical community will likely scrutinize the refutation to ensure its validity. Alpöge has not yet published a detailed paper, so independent verification is pending. A closer look at the same subject appears in AI Breakthroughs in Mathematics: The Future of Human Understanding. Anyone following the wider picture can start with Fable 5.1: US Leads AI Frontier But China Looms. The same theme surfaces in Ai Video: latest articles, which takes a different angle.

If confirmed, this case could lead to more collaborations between mathematicians and AI models, potentially transforming how conjectures are tested and theorems are proven. Context on how this played out before sits in Claude Fails 650 Times at Riemann Hypothesis, Then Succeeds. A companion piece, OpenAI Astra Solves Decade-Old Math Problems, Echoing Deep Blue, approaches the same subject from another angle.