Axiom Math's AI System Formalizes Proof of the '246 Theorem' in Prime Number Theory
Axiom Math's AI system AxiomProver has formally verified the proof of the '246 theorem,' a landmark result from the Polymath8b collaboration on prime gaps. The company says the achievement builds a reusable library for future formalization work and points toward AI verification of software code.
Axiom Math has used its AI system AxiomProver to automatically verify a machine-checkable proof of the "246 theorem," a result in number theory that the company calls its most significant formalization achievement to date.
The 246 theorem states that there are infinitely many prime pairs differing by no more than 246—the closest mathematicians have come to proving the centuries-old twin prime conjecture, which claims infinitely many primes differ by exactly two. The result came out of the Polymath8b collaboration, which included Fields Medalists James Maynard and Terence Tao, building on Yitang Zhang's 2013 breakthrough that first established a finite bound (70 million) between infinitely many prime pairs. Maynard's subsequent work cut that gap to 600 before Polymath8b brought it down to 246.
Formal verification uses software to check a machine-readable version of a mathematical proof, functioning as a rigorous—though not infallible—check on correctness. A recent incident involving a bug in a formal-verification kernel showed the method can be exploited to accept a false, AI-generated proof, according to a report referenced by IEEE Spectrum. Despite that caveat, automated verification remains one of the strongest available guarantees of proof correctness.
A reusable library, not a one-off result
Axiom Math, led by founding mathematician Ken Ono, has used AxiomProver—described as an autonomous, multi-agent system—to formalize several previously unsolved problems this year, according to the company. Sidharth Hariharan, a Carnegie Mellon Ph.D. student now interning at Axiom Math, says the 246 theorem project differs from prior formalization efforts because its components were built for reuse rather than as a single-use proof.
Earlier this year, rival startup Math, Inc. used its Gauss agent to formalize Maryna Viazovska's 2022 Fields Medal-winning sphere-packing proof. Hariharan, who led human efforts on that formalization blueprint, says Axiom Math's work on the 246 theorem is more comprehensive because it produced a general-purpose library of results on prime gaps—published on GitHub as PrimeGapsLib—with the 246 theorem serving as its flagship result.
Why it matters beyond number theory
The number theory techniques underlying the 246 theorem are foundational to cryptography and cybersecurity, meaning the formalized results could eventually help verify systems that protect digital data, according to Axiom Math.
Ono frames the achievement primarily as a proof of concept for a larger goal: verifying AI-generated software code. He argues that if code properties—such as whether an algorithm terminates or produces correct output for all inputs—can be expressed as precise mathematical statements, tools derived from AxiomProver could formally prove those properties, addressing hallucination and bug risks in AI-written code.
"The world is about to run on computer code that nobody has read," Ono said. "AI is here and we can no longer look away—proof formalization is a testbed for solving what I think is the most important challenge we will face from AI."
What this means
This is a research milestone, not a product launch: no new model or API is being released, and Axiom Math has not disclosed pricing, model architecture, or parameter counts for AxiomProver. The genuine significance lies in the strategy shift from one-off formalizations toward reusable proof libraries, which could compound over time the way software libraries do. The stated long-term ambition—using formal verification to check AI-generated code—is speculative and far from demonstrated at scale, but it identifies a real gap: as AI writes more production code, the tools to mathematically certify that code's correctness remain immature. Whether formal methods can scale from pure-math theorems to sprawling, stateful codebases is an open question this work does not yet answer.
Related Articles
Anthropic mathematician Levent Alpöge claims Claude-assisted proof of the Riemann hypothesis, posts it in Lean
Levent Alpöge, a mathematician employed by Anthropic, says he has proven the Riemann hypothesis, first proposed in 1859, working with Claude. He posted the proof coded in Lean, and experts have yet to confirm that it is correct.
AWS releases I/Q, I/R and Isabelle Assistant to connect AI agents to the Isabelle proof checker via MCP
AWS has made three tools in its AutoCorrode verification suite publicly available: I/Q, I/R and Isabelle Assistant. They connect AI agents to the Isabelle proof assistant through the Model Context Protocol, so generated proofs are checked by computer. Amazon Science authors demonstrate the setup using Claude Opus 4.6.
OpenAI outlines four-part approach to combat adversarial distillation
OpenAI describes a layered approach to adversarial distillation. It combines model training, safeguards, detection, and enforcement to protect against capability extraction.
Ai2 launches Theorizer in Asta and open-sources the AI agent that proposes scientific theories
Ai2 launched Theorizer, an AI agent that combines findings from scientific papers to propose explanations for observations, inside its Asta platform. The code is open-source, and Ai2 covers token costs up to a limit in the web interface. A lung cancer case study shows the workflow.
Comments
Loading...