← Back to feed

An Anthropic number theorist and Claude Fable 5 killed one of the most famous open problems in algebraic geometry over the weekend. Levent Alpoge, a former Harvard Junior Fellow now at Anthropic, posted a counterexample to the Jacobian Conjecture to X on July 19, 2026 — during the World Cup Final — in a single message with no accompanying paper, no press release, and no fanfare.

The math community spent the next several hours verifying it. The verification passed.

The Conjecture

The Jacobian Conjecture was formulated by O. H. Keller in 1939. In its simplest framing: if you have a polynomial map F from complex n-space to itself, and that map’s Jacobian determinant is a nonzero constant everywhere, then F must be invertible. The conjecture appeared on Stephen Smale’s 2000 list of 23 unsolved mathematical problems for the 21st century, ranked alongside the Riemann Hypothesis and P vs NP.

It had no known counterexamples. Multiple claimed proofs and disproof attempts over 85 years collapsed on inspection.

The Counterexample

Alpoge’s map operates on C³:

F(x,y,z) = (
  (1+xy)³z + y²(1+xy)(4+3xy),
  y + 3x(1+xy)²z + 3xy²(4+3xy),
  2x - 3x²y - x³z
)

Two properties define the counterexample. First, its Jacobian determinant equals -2 everywhere — nonzero and constant, exactly the condition the Jacobian Conjecture requires an injective map to satisfy. Second, the map is demonstrably not injective: three distinct input points map to the same output.

Input	Output
(0, 0, -1/4)	(-1/4, 0, 0)
(1, -3/2, 13/2)	(-1/4, 0, 0)
(-1, 3/2, 13/2)	(-1/4, 0, 0)

Plugging those coordinates in by hand confirms the collision. No special tools required. A Stanford professor, Joshua Lichtman, publicly walked through the verification within hours and confirmed the result holds.

The conjecture is false. A map can have a nonzero constant Jacobian determinant and still fail to be invertible.

The Human Dimension

The result carries a historical footnote that the mathematics community noted immediately.

Yitang Zhang, the number theorist who in 2013 proved the bounded gaps conjecture and became one of the most celebrated mathematicians of the decade, had his PhD thesis destroyed by the Jacobian Conjecture. His advisor had given him a lemma to build on. That lemma was false, and Zhang’s entire PhD argument collapsed with it. He spent years in near-obscurity before his work on prime gaps was published.

The problem that cost Zhang his thesis fell in a tweet.

What Comes Next

The counterexample operates in C³ (complex 3-space). The Jacobian Conjecture in C² — two variables — carries its own name, the Jacobian plane conjecture, and remains unresolved. Mathematicians moved immediately to ask whether Alpoge’s approach transfers. Initial assessments suggest it doesn’t directly, but the structure of the counterexample opens new lines.

GPT-5.6 Sol, in separate social posts, proposed a potential “repaired” version of the Jacobian Conjecture: one that adds a condition ruling out degeneracy at infinity. Whether that patched variant is provable or simply the next conjecture waiting to fall is now an open question.

What It Means for AI Research

The announcement was not a benchmark run, a curated demo, or a competition submission. It was a working mathematician using a frontier model as a genuine research collaborator to hunt down a specific counterexample to a named, decades-old problem.

Alpoge has the background to evaluate whether what Fable produced was correct. He posted it publicly. Multiple independent mathematicians reproduced and confirmed the arithmetic within hours. The result stands.

Fable 5 has had a turbulent few months — a government-mandated suspension tied to export controls, multiple regulatory rounds, and ongoing alignment review. A clean refutation of a problem from Smale’s list is a different kind of headline.