An Anthropic mathematician has used the company's own large language model to disprove a mathematical conjecture that has resisted resolution for 87 years, delivering one of the most striking examples yet of artificial intelligence advancing pure mathematics.
Levent Alpöge, a mathematician at the AI company Anthropic, announced that he had found a counterexample to the Jacobian conjecture, a famous and long-standing problem in algebraic geometry. He reached the result using Anthropic's Claude Fable 5 model, which had been released to the public only weeks earlier. The discovery, first reported by ScienceDaily and SciTechDaily and drawn from an article originally published in The Conversation, has sent a surge of excitement through the mathematical community. For readers following the latest AI developments, the result underscores how quickly frontier models are moving from parlor tricks into genuinely productive scientific work.
A Formula Small Enough for a Single Post
What makes the breakthrough so startling is its compactness. According to SciTechDaily, the counterexample is a remarkably simple three-dimensional function — a formula short enough to fit inside a single post on X. The AI-assisted counterexample disproves the Jacobian conjecture above two dimensions, while leaving the conjecture's original two-dimensional form open.
That nuance matters. The result does not settle the entire problem, which has stood since the 19th century. Instead, it exposes an unexpected limit: the generalised, higher-dimensional version of the conjecture is false, even though the case mathematicians originally cared about most remains unresolved.
The History Behind the Problem
The Jacobian conjecture concerns polynomial functions — mathematical machines that take in numbers and produce new numbers by following a particular rule. In this setting, numbers can represent points in a space, like coordinates on a map. A function takes those coordinates and transforms them, moving the points to different positions. Mathematicians can examine how smoothly a function rearranges space by calculating its Jacobian determinant. The conjecture essentially asks: if that determinant is a non-zero constant, does the function necessarily have a polynomial inverse?
The problem has a storied history. It was first stated by Czech mathematician Ludwig Kraus in 1884, then generalised to any number of dimensions by German mathematician Ott-Heinrich Keller in 1939. It was considered so important that Fields Medallist Stephen Smale included it in his influential 1998 list of Mathematical Problems for the Next Century.
Over its long life, the conjecture attracted many claimed proofs, including attempts by Beniamino Segre and Wolfgang Gröbner, two celebrated 20th-century mathematicians. None held up. The fact that a counterexample has now surfaced — with help from a chatbot-style model — illustrates how AI is proving as useful for discovering unexpected mathematical objects as it is for constructing proofs.
Why the Higher-Dimensional Case Fell
The key distinction is between two and three (or more) dimensions. The original two-dimensional Jacobian conjecture, the version that has occupied algebraic geometers for generations, is still open. What Alpöge and Claude Fable 5 found is a counterexample in three dimensions, which is enough to knock down the generalised conjecture that Keller formulated in 1939.
That is a meaningful result in its own right. Many researchers had suspected that if the conjecture were true at all, it would be true everywhere. A clean counterexample in dimension three narrows the space of possibilities and reframes the remaining question: is two dimensions genuinely special, or is there a counterexample lurking there too?
The brevity of the counterexample is itself remarkable. Fortune reported that mathematicians are grappling with a "very rapid and very unsettling change" as AI tools crack century-old problems one after another. A tiny formula toppling a problem endorsed by a Fields Medallist is exactly the kind of development that fuels that unease — and that excitement.
AI's Growing Role in Mathematics
Mathematicians working with large language models have recently produced several notable results, and the Jacobian conjecture counterexample stands apart from many earlier advances. Rather than grinding through a brute-force search, the collaboration between Alpöge and Claude Fable 5 produced a concise, human-checkable answer — the kind of output that mathematicians can verify and build upon.
The discovery also highlights a broader pattern. Frontier models are increasingly capable of exploring vast combinatorial spaces and proposing candidate objects that a human might never stumble across. When those candidates are short enough to verify by hand, the combination of AI breadth and human judgment becomes potent.
What It Means for the Future
What this means for the future of mathematics — and for human mathematicians — remains to be seen, as SciTechDaily noted. The counterexample does not render mathematicians obsolete; if anything, it sharpens the open question and hands researchers a precise new target. But it does reinforce that AI has moved beyond summarising textbooks and into the heart of creative mathematical work.
The result is also a high-profile validation for Anthropic. Claude Fable 5, released to the public only weeks before the discovery, was used by one of the company's own mathematicians to resolve a problem that had thwarted some of the greatest minds of the past century. For a company racing to demonstrate that its models can reason at the frontier, few endorsements are more credible than a 1939 conjecture finally cracking.
Stay Ahead of AI
The pace of AI-driven discovery is accelerating across science, mathematics, and industry. Bookmark our breaking AI news desk for continuous coverage of model releases, research breakthroughs, and policy shifts.
Read more AI news →
