AI labs say their models just solved famous math problems. Here's what's real so far
This week AI went from writing code and emails to claiming breakthroughs that would normally earn a mathematician a career's worth of recognition. OpenAI says its models made three new discoveries in math, computer science and physics. Within hours, Anthropic said its model had reached one of the same results. Here's what was claimed, what experts say, and why it matters even if you never touch a proof.
What was announced
- Math: Harmonic analysis. Claim: Proved the local smoothing conjecture in every spatial dimension. Before: It had only been proved in three dimensions. Reaction: Rochester mathematician Alex Iosevich said OpenAI appeared to have cracked it.
- Computer science: Complexity theory. Claim: Refuted the Exponential Time Hypothesis (ETH). Also: Anthropic released its own refutation hours later. Reaction: MIT's Ryan Williams: major if it holds; experts still checking.
- Physics: Quantum field theory. Claim: A mathematically rigorous proof about quantum chromodynamics, the theory of quarks and gluons. Release: Physicists are deciding how to release and interpret it. Status: Few details public yet.
Sources: Al Jazeera, Oct 27; TechCrunch, Oct 27. Claims are the company's; experts are still checking.
The Exponential Time Hypothesis, in plain English
Some puzzles are easy to check but hard to solve. A finished Sudoku takes seconds to verify; solving a giant one can take forever. ETH, first published in 1999 by Russell Impagliazzo and Ramamohan Paturi, says that for a certain class of hard logic puzzles, even the best possible method gets exponentially slower as the puzzle grows.
Computer scientists have used that assumption in countless papers. The labs say their models found methods that beat that limit. If that holds up, a lot of research built on ETH may need revisiting, TechCrunch reports.
What the experts say
- Not verified yet. Williams, who is organizing the computer science community's review, said parts of the proof look routine and a few look unusual enough to need community-wide work. He doesn't think OpenAI's method is likely the best possible.
- Was it new? Anthropic mathematician Levent Alpöge suggested OpenAI's solutions drew on existing work, shortly before Anthropic published its own refutation.
- No short-term impact. Williams told Al Jazeera a confirmed result would not have near-term real-world applications.
- What about encryption? Asked whether the result breaks cryptography, OpenAI's science lead Dan Roberts reportedly said the team checked and was reasonably confident it doesn't.
- Who sees it first. OpenAI used outside expert groups to release the work and acknowledged that picking those groups gives some people early access.
What it means for you
- AI is moving from assistant to researcher. The same kind of models you use for writing and coding are now producing results experts take seriously. Expect more "AI discovered X" headlines.
- Treat every headline as "claimed" until checked. Even here, the right status is "experts are verifying." The careful reaction is the correct one.
- Your passwords and bank encryption aren't suddenly broken. By OpenAI's own check, this doesn't break cryptography, and it isn't a practical tool yet.
- The skill that matters is judgment. As AI produces more answers, knowing how to check them becomes more valuable, at work and in life. The free AI library on this site has prompts for asking AI to show its sources and reasoning.
Disclosure: this site's posts are drafted with AI help. This one was drafted with Claude, made by Anthropic, one of the labs in this story. It sticks to published reporting and labels every claim.
Sources
- Frontier AI labs say their models refuted the Exponential Time Hypothesis. What does that mean?, TechCrunch, Tim Fernholz, Oct 27, 2026
- AI makes discoveries in mathematics, computer science, physics, OpenAI says, Al Jazeera, Oct 27, 2026