OpenAI's AI Math Surge Sparks Mathematician Boycott and Crypto Security Alarms

On October 6, 2026, OpenAI released 722 AI-generated mathematical manuscripts organized into 372 result families via a public GitHub repository. The papers, produced by an unreleased internal frontier model tested on roughly 4,000 problems, claim advances on longstanding challenges including a zero-free region for the Riemann zeta function and Dirichlet L-functions at Re(s) > 7/8—termed a quasi-Riemann hypothesis—alongside progress on the Hodge conjecture for CM abelian varieties and other Millennium Prize Problems. Many include Lean formalizations for verification, with average compute costs equivalent to three hours of ChatGPT Pro reasoning.[1][2]

Rutgers mathematician Alex Kontorovich described the quasi-Riemann result as potentially Fields Medal-worthy if achieved by a human. However, the scale and proprietary nature of the release prompted backlash. The Association for Human Mathematics (AHM) issued a statement urging researchers to boycott OpenAI, arguing the company disregarded scientific norms by prioritizing proprietary AI output over human understanding and releasing results en masse without broader consultation. AHM, separate from advisory groups, emphasized that mathematicians did not request this work and called it a demonstration of power rather than scholarship.[3][4]

Beyond mathematics, the release has amplified concerns in cryptography. Bitcoin researcher Justin Drake warned of 'bunker mode' preparations for large crypto holders, suggesting AI-driven mathematical breakthroughs could threaten elliptic-curve cryptography (ECDSA) underpinning wallets and smart contracts sooner than quantum computing threats materialize. This echoes earlier discussions following OpenAI's Navier-Stokes claims, highlighting how accelerated AI math could automate attacks on cryptographic assumptions like prime distributions or discrete logarithms.[5]

Regulatory implications loom: if AI can systematically erode the mathematical foundations of digital security, frameworks for AI oversight in critical infrastructure, export controls on advanced models, and standards for verifiable proofs in finance may need urgent evolution. The episode underscores tensions between rapid AI progress and established epistemic practices in mathematics and security.