OpenAI says its artificial intelligence models have disproved a 30-year-old assumption in computer science known as the exponential time hypothesis, or ETH. The claim, made in a company blog post on October 26, would mean an algorithm can solve certain logical puzzles faster than the assumed speed limit allows.
The ETH is a conjecture about how quickly hard logical problems can be solved. Many results across computer science rest on it, because it sets boundaries on what computers can realistically compute. If the assumption is wrong, a number of those downstream conclusions would need to be revisited. Researchers have been relying on the hypothesis for decades, which is why a challenge to it draws close attention.
The company also made several other claims in the same announcement. It said its models produced a proof of a 1991 conjecture about wave focusing, and that certain quantum field theories are mathematically coherent, with one part of that work, concerning Yang–Mills theory, not yet released. None of these additional claims has been independently verified.
The ETH result was checked with Lean, a proof-checking software system that confirms whether a mathematical argument is logically valid. A mathematician who reviewed it said the proof appears sound. An independent group that includes computer scientists Russell Impagliazzo and Ramamohan Paturi, both established figures in complexity theory, is now examining the findings.
Even if the ETH is overturned, the result would not settle one of computer science's central open questions. Disproving the hypothesis does not resolve the P versus NP problem, which asks whether every problem whose solution can be quickly checked can also be quickly solved. Researchers also said they do not yet understand why the algorithm works.
Some academics have raised concerns about results of this kind being announced by a private company rather than passing first through peer review. OpenAI has said it is exploring ways for academic researchers to verify its work independently.
We rewrote this story on 4 November 2026 after an editorial review.
Credit. Based on reporting by Chris Stokel-Walker for New Scientist. Is this your work? Write to us.
490





