Fields medalist Terence Tao, an influential voice on the use of AI in mathematical research, has published a post titled "A severe misalignment of AI in mathematics," in which he raises concerns about a troubling gap between OpenAI's public claims and the actual practice of mathematics. The same day, The Economist published a piece headlined "Top mathematicians are outraged by OpenAI's methods," suggesting the unease extends well beyond a single critic and touches a meaningful portion of the field.
The dispute follows a string of announcements OpenAI has made since 2025 regarding the performance of its reasoning models on high-level problems, including claimed results at the level of the International Mathematical Olympiad and on unsolved research questions. Those announcements had already sparked debate over how AI capabilities in mathematics are measured, verified, and communicated to the public, often outside the peer-review channels the discipline traditionally relies on.
While the full substance of the grievances is not yet detailed, the headlines of both pieces point toward criticism of OpenAI's methodology in establishing or presenting mathematical results, rather than a dispute over the raw technical abilities of the models. Such controversies touch on sensitive issues for the field: the traceability of proofs, the distinction between automated problem-solving and genuine understanding, and the risk that training data has been contaminated with competition or benchmark statements.
The episode reflects a broader tension between AI labs eager to showcase impressive breakthroughs and a scientific community committed to rigorous verification standards. That the criticism comes from a figure as prominent as Terence Tao gives it particular weight, and it is likely to shape how future mathematical performance claims from major AI labs are scrutinized going forward.