25 Fields Medalists Warn AI Turns Math Into a Speed Contest

Twenty-five Fields Medalists, including Terence Tao and Yu Deng, have issued a joint statement warning of a severe misalignment between the goals of artificial intelligence developers and the mathematical research community. Titled A Severe Misalignment of AI in Mathematics, the declaration emphasizes that while AI companies are increasingly benchmarking model capabilities by rapidly solving famous mathematical conjectures, the academic community remains focused on deep understanding, rigorous verification, and long-term knowledge integration. The statement follows a series of high-profile claims by OpenAI regarding its internal models. In September, the company announced that an undisclosed AI system, supported by approximately 10,000 parallel agents, generated a purported proof for the Navier Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. The computational effort reportedly involved 130 billion output tokens and 2.7 million inter-agent messages, followed by Lean-based formal verification. OpenAI further indicated progress on another Millennium problem, heightening concerns about the accelerating pace of AI-generated mathematical results. According to the signatories, this technological acceleration threatens to disrupt established research practices. The traditional mathematical process requires years of peer review, simplification, and pedagogical adaptation to transform complex proofs into usable knowledge. AI systems, optimized for speed and output volume, compress this lifecycle into isolated breakthroughs, potentially bypassing the iterative reasoning and conceptual framing that define mathematical training. Tao illustrated this dynamic with a waterfall versus helicopter analogy, noting that while AI can transport researchers directly to a solution, it does not automatically generate the underlying conceptual maps required for genuine understanding. The declaration also addresses downstream impacts on research integrity and academic development. Rapid AI dissemination may outpace the verification process, creating disputes over priority, attribution, and data handling. Furthermore, overreliance on automated theorem generation risks undermining the mentorship structures that cultivate problem formulation, intuition, and error analysis. Yu Deng recently noted that while AI has not surpassed human capability in mathematics, its rapid iteration has introduced a palpable sense of urgency within the field. The authors do not oppose AI integration into mathematical research but stress that technological progress must align with human cognitive and educational timelines. As AI continues to compress discovery cycles, the community must establish frameworks for validation, authorship, and pedagogy that prioritize depth over speed. The declaration calls for a recalibration of how progress is measured, ensuring that AI serves as an accelerant for insight rather than a substitute for the deliberate practice that sustains mathematical discovery.