On July 20, Anthropic researcher Levent Alpoge announced a sensational result: the Claude Fable 5 language model found a counterexample to the Jacobian conjecture. This problem, formulated by German mathematician Ott-Heinrich Keller in 1939, had remained one of the central unsolved problems in algebraic geometry for decades.
The Essence of the Refutation
The Jacobian conjecture states that if a multidimensional polynomial mapping is locally invertible (i.e., its Jacobian is nowhere zero), then it is globally invertible. In simpler terms, if the formula "locally" does not lose information, then it can be inverted "globally." However, Claude Fable 5 constructed a counterexample—a specific formula that passes the local test at every point but maps three different sets of numbers to the same result. This makes complete recovery of the original data impossible, thereby refuting the conjecture.
Notably, the question about the problem was posed by a friend of Alpoge, and the model itself worked on it during the World Cup final. The result is already supported by references to calculations in Wolfram Alpha, allowing anyone to verify it independently.
Context and Significance
The Jacobian conjecture is known not only for its complexity but also for the abundance of erroneous proofs. Mathematician Qiaochu Yuan noted that this is the most famous open problem solved by a language model. He cited a 2008 paper by T. T. Moh, which mentioned that Beniamino Segre published three incorrect proofs, Claude Chevalley accepted a flawed proof as correct, and Igor Shafarevich used the conjecture as a proven theorem.
Mathematician Jared Duker Lichtman called the result "outstanding," emphasizing that a special case of the conjecture was the topic of Ethan Zhang's dissertation—the future author of a breakthrough in prime number theory. Although the result has not yet undergone peer review, its key feature is verifiability: anyone can plug in the formula and recalculate.
AI vs. Open Problems
This is not the first time AI has solved mathematical problems. In October 2025, OpenAI's vice president claimed that GPT-5 solved ten Erdős problems, but this was not confirmed—the model found already existing solutions. In January 2026, GPT-5.2 Pro solved Erdős problem No. 397, and in May, an OpenAI model refuted Erdős's 1946 conjecture on unit distances. Google DeepMind and the Star Fleet Math project also presented solutions using AI.
My analysis: The refutation of the Jacobian conjecture is a significant step for AI in mathematics. However, it is important to remember that such results are just the beginning. For now, models are picking the "low-hanging fruit," as Terence Tao noted. But Claude Fable 5's ability to create a verifiable counterexample to one of the most famous open problems demonstrates that the boundaries of AI capabilities are expanding. The cryptocurrency and blockchain market, where data accuracy and verifiability are critical, can learn from this story: automation of complex system analysis is becoming a reality.