Terence Tao warns AI could trigger math's biggest crisis since Gödel
Fields Medalist Terence Tao argues in a new essay that AI is about to force mathematics to confront questions it has spent a century avoiding — and the reckoning could be as destabilizing as Gödel's incompleteness theorems.
Tao's central claim is that AI will soon handle a meaningful fraction of research-level math, and the field is nowhere near ready for the consequences. Writing for the 2026 International Congress of Mathematicians, Tao draws a direct line to the foundational crisis of 1900–1930, when Russell's paradox and Gödel's theorems forced mathematicians to make their implicit assumptions explicit. That crisis produced the rigorous framework that has held for a century. Today, Tao argues, the stress test has shifted from mathematical truth to "the largely implicit framework of mathematical values and practices" — what counts as a contribution, what gets rewarded, and whether a machine can be said to have done the work. His working hypothesis: AI tools will "reasonably soon" become capable of performing research-level mathematical tasks with acceptable quality and cost. As evidence, he points to the First-Proof Project, where ten never-published problems were tested against four AI systems under controlled conditions. Seven of the ten received at least one passing grade from at least one system, at costs in the tens to hundreds of dollars per problem. The danger, Tao writes, is Goodhart's law in action: when AI makes proof generation cheap, the metric of "producing proofs" stops being a proxy for understanding. The field could shift from proof scarcity to proof abundance, with AI-generated proofs piling up faster than anyone can check or absorb — and the financial incentives of the AI industry will reward exactly the kind of benchmarkable wins that mathematicians have long used as stand-ins for deeper goals.
Callosum raises $100 million seed to route AI tasks to the right models and chips. The London-based startup, backed by Atomico and the UK Sovereign AI Fund, has built software that matches AI workloads with the optimal model and hardware combination — a routing layer for an increasingly fragmented inference market. As enterprises deploy a mix of frontier models, open-weight alternatives, and specialized chips, the problem of "which model runs this task best on which hardware" is becoming a real operational headache. Callosum's pitch is that this matching problem is tractable with the right metadata and benchmarking infrastructure, and that a $100 million seed gives them the runway to build it before the big cloud providers close the gap. The round is one of the largest seed-stage raises in European AI this year, signaling that investors see model routing as infrastructure, not a feature.
Google DeepMind launches Backstory, an experimental AI image authentication tool for journalists and fact-checkers. The tool is designed to help verify whether AI-generated or manipulated images are authentic — a direct response to the growing flood of synthetic media in newsrooms. Backstory is being offered in limited testing to journalists, researchers, and other fact-checkers, according to Nieman Lab. The tool sits at the intersection of DeepMind's research capabilities and the practical needs of media organizations struggling to keep up with AI-generated misinformation. It's a notable move from Google, which has been criticized for slow action on AI-generated content in its own search results — now one of its research labs is building tools specifically for the people trying to hold that content accountable.
What to watch: Tao's essay is a gauntlet thrown at the math community — will it change how departments evaluate AI-assisted research, or will the incentive structures he describes prove too durable to disrupt?
Is a $100 million seed round for model routing a sign that the AI stack is fragmenting faster than anyone expected — or is it too early for a routing layer to matter? Tell us in the comments.
