Terence Tao Warns AI Is Outpacing Math’s Hardest Problems

Fields medalist Terence Tao has warned that artificial intelligence is creating a new race in mathematics, with systems able to flatten difficult problems as soon as researchers begin working on them and competition between OpenAI and Anthropic illustrating the pace of change.

Tao’s warning centers on the speed at which advanced AI systems are changing the relationship between researchers and difficult mathematical questions. Rather than simply assisting with routine calculations or checking established work, the systems described in the report can rapidly reduce the difficulty of problems that researchers have only just begun to investigate.

A shrinking window for mathematical discovery

The development creates a problem for the traditional pace of mathematical research. Hard problems have historically offered researchers long periods in which to develop new approaches, test conjectures and establish results. Tao’s observation suggests that this window is becoming shorter: a question may remain difficult for a human researcher at the moment it is identified, yet become substantially more tractable once an advanced AI system is directed toward it.

The concern is not that mathematics is running out of questions. Instead, it is that the most challenging questions may be solved faster than the field can generate replacements of comparable difficulty. That distinction is central to Tao’s warning. The issue is the rate at which frontier systems are consuming difficult, unresolved work, not an assertion that mathematical research itself is nearing an endpoint.

OpenAI and Anthropic highlight the competition

The report points to a real race between OpenAI and Anthropic as evidence of the trend. Their competition is presented as a practical demonstration of how quickly capabilities are advancing, rather than as a separate development in the story. As companies compete to build more capable systems, the ability to address difficult mathematical problems becomes part of a broader contest over reasoning performance.

That competition matters because improvements can have effects beyond a single proof or research project. When systems become better at breaking down complex mathematical tasks, researchers may encounter a rapidly changing baseline for what counts as difficult, novel or time-consuming work. A problem that once required a sustained human effort could be approached differently if a model can identify a useful path almost immediately.

What Tao’s warning means for researchers

The immediate implication is a shift in how mathematicians may need to evaluate their work. AI systems could accelerate the search for solutions, but the report’s emphasis is on the pressure created by that acceleration. Researchers may have less time to explore a problem independently before automated systems produce a decisive advance.

This could also change the value of different stages of mathematical research. Finding a question, defining it precisely and establishing why it matters may become as important as solving it. If AI can quickly flatten a problem once it has been formulated, the ability to identify meaningful questions could become a more persistent source of human contribution.

• Faster solutions: Difficult problems may become manageable shortly after researchers begin addressing them.
• Shorter research cycles: The time between recognizing a challenge and resolving it could continue to contract.
• Greater emphasis on problem selection: Researchers may place more value on choosing questions whose significance extends beyond their technical difficulty.
• Competitive pressure: Rivalry among leading AI companies could accelerate improvements in mathematical reasoning.

A warning about the pace of change

Tao’s comments frame AI progress as a race between capability development and the production of new mathematical challenges. The concern is therefore about sustainability and direction: if systems solve the hardest available problems faster than mathematicians can replace them, the field may need to reconsider how it defines progress and originality.

For 2026, the warning offers a clear measure of AI’s expanding reach. The significance lies not only in whether a system can solve a particular problem, but in whether its speed changes the research environment around that problem. Tao’s assessment suggests that the frontier is moving quickly enough for mathematics to feel the effects in real time.

The result is a more urgent debate over the role of human researchers in an era when difficult questions may no longer remain difficult for long.