On July 20, 2026, mathematician Levent Alpöge posted a 216-character counterexample to the Jacobian Conjecture — an open problem since 1939 — discovered with Claude Fable 5 during the FIFA World Cup final. The result has been independently verified and definitively closes the conjecture for dimensions n≥3. DrafterDaily explains what happened, what it means for AI's role in research, and what the result still can't tell us.
On the evening of July 19, 2026, Spain beat Argentina 1–0 in the FIFA World Cup final. While billions of people watched the match, Levent Alpöge — a mathematician who works with Anthropic — was doing something else with part of his attention: he was asking Claude Fable 5 to help him think about a polynomial map. The next morning, he posted the result on X. The Jacobian Conjecture, an open problem in pure mathematics since 1939, had been disproved. The counterexample was 216 characters long.
The Jacobian Conjecture was proposed in 1939 and asks a deceptively simple question about polynomial maps — the kind of mathematical functions that take input values and transform them using polynomial expressions. Specifically, it asks: if you have a polynomial map from complex n-dimensional space to itself, and the Jacobian determinant of that map is a non-zero constant everywhere, does it follow that the map is globally invertible? In intuitive terms: if you can locally undo the transformation at every single point, can you always undo it globally?
For 87 years, mathematicians assumed the answer was yes — or at least could not prove it was no. The conjecture appeared in hundreds of papers, motivated research across algebraic geometry and commutative algebra, and resisted every proof attempt. It remained on the list of major open problems in mathematics not because it seemed obviously false, but because nobody could find a counterexample and nobody could prove it true. That is the definition of a hard problem.
The counterexample Alpöge posted is a polynomial map from C³ to C³ — from three-dimensional complex space to itself. Its Jacobian determinant is identically equal to −2 everywhere. By the terms of the Jacobian Conjecture, a map with a constant, non-zero Jacobian determinant should be globally invertible. But this one is not. Alpöge's counterexample sends three distinct input points to the same output point, definitively proving that global invertibility does not follow from a constant Jacobian. The conjecture is false — at least for n≥3. The case n=2 (two-dimensional complex space) remains open and may still be true.
In his post, Alpöge credited two people: Akhil Mathew, a fellow mathematician, for asking the right question that led him to look in this direction, and Claude Fable 5 for doing the exploratory work during the World Cup final. This attribution matters. The counterexample was not produced by typing 'disprove the Jacobian Conjecture' into a chat window. It emerged from a specific question that a human mathematician had the expertise and instinct to ask. Fable 5 explored the combinatorial and algebraic space that the question opened up. Alpöge recognized the result.
Within hours of the post, mathematicians worldwide checked the result independently. The two key claims — that the Jacobian determinant of the map is identically −2, and that three distinct inputs map to the same output — can be verified using standard computer algebra systems, and they check out. The result stands. It has been described as the hardest mathematical conjecture ever resolved with AI assistance.
The framing that 'AI solved an 87-year-old math problem' is technically accurate and practically misleading. A more useful framing: a mathematician using AI as a collaborator was able to explore a region of mathematical space that he might not have explored efficiently alone, and found something there that neither he nor the field had found in eight and a half decades of trying. The leverage was the question — Akhil Mathew's question, posed to Alpöge, which gave Claude Fable 5 a defined search space to work within.
This pattern — human expertise defines the problem, AI explores the space, human validates the output — is emerging across several fields simultaneously. It is not AI replacing researchers. It is researchers with AI covering more intellectual ground per unit of time than was previously possible. The Jacobian result is notable precisely because it happened at the frontier of pure mathematics, where the search space is vast, the landmarks are subtle, and human intuition has historically been the only reliable guide. AI is now useful in that space.
“The contribution that unlocked 87 years of mathematical deadlock was not computational power. It was a mathematician asking the right question.”
There is a limit to what this result demonstrates, and it is important. Claude Fable 5 produced the counterexample. It cannot explain, in any mathematically meaningful sense, why the counterexample works — what property of the map causes the invertibility to fail, what intuition would have led a human directly to this region of the search space, or whether there is a deeper structural reason that the conjecture fails for n≥3 but may hold for n=2. The counterexample is a fact. The understanding of that fact — the mathematical intuition behind it — is still a human project.
This matters because mathematics is not primarily in the business of accumulating facts. It is in the business of understanding. A proof is not just a certificate that something is true; it is an explanation of why it is true in a way that illuminates the surrounding mathematical landscape. The Jacobian result tells us the conjecture is false. The mathematical community will now spend months or years building the understanding of why, using the counterexample as a starting point. That work will be done by mathematicians.
Key takeaway: AI found a 216-character fact that eluded 87 years of human mathematical effort. AI cannot tell us what that fact means. Those are different capabilities, and confusing them is the central mistake in most coverage of AI and mathematics.
The Jacobian Conjecture is now settled for n≥3: it is false. But the two-dimensional case — polynomial maps from C² to C² — remains open. Alpöge's counterexample does not apply to two dimensions, and the n=2 case may still be true. This is not a footnote. A proof of the two-dimensional case would have significant implications in algebraic geometry and could illuminate the structure of polynomial automorphisms in ways that the n≥3 disproof does not. The conjecture lives on in its most mathematically interesting form.
For mathematics, the immediate task is understanding the structure of Alpöge's counterexample deeply enough to inform the remaining n=2 problem and related questions in polynomial automorphism theory. For AI research, the Jacobian result joins a growing list of cases — including recent AlphaProof work on International Mathematical Olympiad problems — that suggest AI is becoming a useful tool at the highest levels of abstract reasoning, not just applied domains. For anyone thinking about the future of knowledge work, the result is a concrete data point: the problems AI helps solve are getting harder, and the workflow that produced this result — human question, AI exploration, human recognition — is worth understanding now, not later.
The Jacobian Conjecture asked whether a polynomial map with a constant, non-zero Jacobian determinant must be globally invertible. It was posed in 1939 and resisted resolution for 87 years because no one could find a counterexample and no one could prove it true. The search space for counterexamples is enormous and subtle, which is precisely why AI assistance — capable of exploring combinatorial and algebraic spaces rapidly — was useful in finding one.
The analytical briefing on AI, technology, and business — every morning.
Claude Fable 5.1 costs exactly what Fable 5 cost per token. The 25-to-45% saving Anthropic advertises comes from one repriced line item — cached input, now billed at 2.5% of list instead of 10%. That is a discount you only collect if you keep the agent running.
Anthropic was not breached. The malware was already on the customer's machine, and it took a session cookie rather than a password — which is why two-factor authentication did nothing and why server-side revocation is the only lever the vendor has.
OpenAI's incident report on the Hugging Face breach leads on a security failure. The remarkable part is a verbatim chain-of-thought in which a model identifies its action as unauthorised and proceeds anyway — and an escape route that was a package manager, not a superintelligence.