A Watershed Moment in Algebraic Geometry
In a development that has sent tremors through the global mathematical community, Harvard mathematician Levent Alpöge announced on July 20, 2026, that he has identified a potential counterexample to the 87-year-old Jacobian Conjecture. The discovery was facilitated by Anthropic’s experimental AI model, Fable 5, and consists of a remarkably concise, 216-character string of equations. While the finding currently awaits formal peer review, it represents a significant milestone in the rapidly evolving intersection of artificial intelligence and advanced theoretical research.
The Jacobian Conjecture, first proposed by German mathematician Eduard Ott-Heinrich Keller in 1939, has long been considered one of the most stubborn problems in algebraic geometry. It concerns the nature of polynomial transformations and has remained a cornerstone challenge for mathematicians for nearly a century. The fact that an AI model identified a potential flaw in a problem of this magnitude—one that has resisted human efforts for decades—has ignited an intense debate regarding the future of mathematical methodology.
The Accelerating Pace of AI-Driven Discovery
The breakthrough arrives at a time of unprecedented AI activity in mathematics. In May 2026, an OpenAI reasoning model successfully overturned Paul Erdős’s 1946 “Unit Distance Conjecture,” followed closely by Google DeepMind’s AlphaProof Nexus, which independently solved several open Erdős problems. These events have moved the goalposts for what is considered “autonomous” research. Unlike previous AI efforts that acted as glorified search engines for existing literature, these recent models are demonstrating the capacity to generate novel proofs and identify counterexamples that human researchers had overlooked.
Li Xinyi, an associate professor at Peking University’s International Center for Mathematical Research, notes that the success of these models is not solely due to raw computational power. Instead, it reflects a strategic shift: AI is proving capable of bridging disciplinary silos by applying tools from disparate fields—such as number theory—to geometric problems, a task that often eludes human specialists constrained by traditional academic boundaries.
The Leiden Manifesto and the Fight for Autonomy
The sudden influx of AI-generated proofs has prompted a collective defensive posture among leading mathematicians. On June 2, 2026, 16 researchers from 15 universities released the “Leiden AI and Mathematics Manifesto,” which has since received formal backing from the International Mathematical Union (IMU). Signatories, including Fields Medalists Terence Tao and Peter Scholze, argue that the current trajectory poses systemic risks to the field.
The manifesto highlights several critical concerns, most notably the “indigestion” of the peer-review system. As AI models churn out proofs at high speeds, the limited pool of human experts qualified to verify these results faces an impossible workload. This creates a risk of “AI-generated misinformation” entering the academic record. Furthermore, there are growing anxieties regarding attribution and the influence of commercial entities. The manifesto asserts that the mathematical community must retain the authority to decide which problems are worth solving, rather than allowing corporate AI development agendas to dictate the direction of fundamental research.
Human Verification as the Final Frontier
Despite the speed of AI discovery, human verification remains the essential “last line of defense.” When OpenAI overturned the Unit Distance Conjecture, nine top mathematicians were required to manually review and refine the proof to ensure its accuracy. Similarly, while the Fable 5 counterexample to the Jacobian Conjecture is currently being scrutinized by independent researchers, it has not yet been accepted as definitive. As Li Xinyi observes, while AI is currently proficient at “filling in the convex hull” of known mathematical knowledge, it has yet to prove it can reliably explore the truly unknown frontiers of the field without human oversight.