Terence Tao Warns AI Giants Turn Math Proofs Into Toxic Waste

Fields Medal-winning mathematician Terence Tao has criticized major AI companies for treating unsolved mathematics as a benchmark race, arguing that machine-generated proofs can burden researchers with large volumes of poorly understood results that he describes as toxic waste. In a recent interview with The Observer, Tao said companies such as OpenAI and Anthropic are increasingly focused on how many hard problems their models can solve, rather than on whether the resulting proofs are clear, elegant, or useful to the mathematical community. His comments follow OpenAI’s release of results covering 722 previously unresolved problems, a move that he said creates a new challenge for mathematicians: validating, interpreting, and integrating AI-generated material into existing knowledge. Tao also described intensifying competition between leading AI labs. He alleged that after reports Anthropic researchers might be close to progress on one of the Millennium Prize Problems, including questions related to the Navier-Stokes equations, OpenAI raised the priority of its mathematical work, deployed a large number of AI agents, and redirected compute, reportedly delaying new user access to some services. OpenAI denied the most dramatic aspects of that account, including claims of a top-alert internal response, suspended signups, and the funding scale Tao cited. A separate OpenAI result linked to Navier-Stokes has itself faced expert scrutiny and possible errors. The broader concern, Tao argued, is that mathematics is not only about producing correct statements. In the traditional model, a proof contributes new concepts, methods and connections through peer review, simplification and discussion. AI systems may now generate many formal arguments that are technically checkable but hard for humans to comprehend. Tao pointed to Anthropic’s efforts to pay mathematicians to review and organize AI findings, while warning that reliance on proprietary frontier models could shift control over research agendas and publication opportunities. Tao is not anti-AI; he has used AI tools himself and helped advocate formal verification methods. In a Nature comment, he called for a Maths 2.0 era in which the field places less emphasis on problem-solving as a leaderboard and more emphasis on understanding, collaboration, verification and explanatory value. His central message is that AI can accelerate discovery, but it cannot replace the mathematical work of turning answers into knowledge.