Key Takeaways
Anthropic's Claude Fable 5 produced a counterexample to the Jacobian conjecture in 216 characters, a feat that strengthens the case for pouring capital into AI at crypto's expense.
Anthropic's Claude Fable 5 disproved the Jacobian conjecture, an open problem in algebraic geometry dating to 1939, by producing a concrete counterexample that mathematicians verified by hand within a day. The result, posted on X by Anthropic number theorist Levent Alpöge on Sunday, marks the first time an AI has cracked a problem prominent enough for a Fields medalist to have heard of it.
"This is exactly the kind of thing I would expect AI to be able to do," Andrew Blumberg, a mathematics and computer science professor at Columbia University who sits on the board of the First Proof project testing frontier models on research-level math, said. "If there was a counterexample that was concise and easy to state that people haven't found because it's a pain to search through all this stuff, AI will find it."
The counterexample is a polynomial map from three-dimensional complex space to itself that passes the Jacobian determinant test — the standard check for reversibility — yet still cannot be inverted because three different inputs produce the same output. The Jacobian conjecture had held that any map passing that check must be reversible. One valid counterexample is all it takes to sink an 87-year-old conjecture. Alpöge credited "fable" — the public version of Claude Mythos Preview, which Anthropic had previously described as too capable for public release — for working through the World Cup final to find it.
The breakthrough lands days after China's Moonshot AI released Kimi K3, a model that beat Claude and GPT in coding benchmarks and sent bitcoin down sharply last Friday. Bitcoin has spent months trading on AI narratives, moving with chipmakers and memory stocks rather than any crypto-specific catalyst. Bank of America said Kimi K3 reinforces its bullish thesis on memory and Micron Technology, sending shares of Micron up about 6% premarket Tuesday. The connection is partly structural: bitcoin miners have rebuilt themselves into AI data-center operators, so their fortunes now rise and fall with demand for computing power.
What the Jacobian conjecture was, and why it fell
Think of a mathematical machine that takes two numbers and returns two new numbers using only addition and multiplication. The question, first posed by Ott-Heinrich Keller in 1939, was whether such a machine can always be run backward: given only the output, can the original inputs be recovered every time? Mathematicians had a quick test — the Jacobian determinant — to check whether a machine looked reversible. The conjecture said that if a machine passed that test, it must be reversible.
For 87 years, nobody could prove it true, and nobody could find a counterexample. Fable 5 found one: a map that passes the test but still cannot be reversed, because three distinct inputs all yield the same output. The counterexample is a plain list of polynomials that any mathematician can verify by hand. Finding it was the needle-in-a-haystack part, and that is where the machine excelled.
The reaction among mathematicians was swift. Timothy Gowers, a Fields medalist, called it the first time an AI had solved a problem outside his own area that was big enough for him to have heard of it. Daniel Litt posted at 2 a.m. that he could not stop laughing. The Stanford number theorist Jared Duker Lichtman noted the historical irony: the conjecture once helped sink the PhD of Yitang Zhang, who later became famous for work on prime gaps. An AI settling it in a tweet, he said, was poetic.
Not everyone was impressed. Blumberg contrasted the counterexample with OpenAI's disproof of the Erdős unit distance conjecture in May, which mathematicians could unpack and build on. "This counterexample tells us essentially nothing," he said. "There are a lot of polynomials, and it's hard for people to check them all, but it's not hard for machines."
Why AI capability leaps matter for capital allocation
Each result like Fable's finding strengthens the case for pouring capital into AI, and poses a difficult question for crypto investors: Why hold a token that trades as a sidecar to the AI cycle when someone can own the vehicle itself? The speculative money and investor attention that once chased crypto are now chasing compute, chips and model builders, and every leap in what these systems can do widens that appeal.
The AI capability curve is steep, and the steeper it gets, the more of the market's risk appetite it draws away from everything else, crypto included. Bitcoin miners-turned-AI operators face a direct trade-off: every dollar of computing power allocated to AI inference is a dollar not spent on securing the bitcoin network. Galaxy Digital this week set up a $5 million fund to help shield bitcoin against quantum computing threats, a recognition that the same AI advances driving model breakthroughs also accelerate the timeline for cryptographic risks.
Alpöge hinted the story may not end in destruction. Unpicking his own counterexample, he wrote that it seems to show a glimmer of a positive result hiding inside. A full write-up will follow. OpenAI researcher Aaron Lou said an internal version of the company's Codex model found essentially the same counterexample on its own. And one developer has already used GPT-5.6 to turn the single counterexample into an infinite family, one for every whole number above two — suggesting the machines that broke the old rule are already writing the next one.
This article is for informational purposes only and does not constitute investment advice.
