Frontier AI labs say their models refuted the Exponential Time Hypothesis. What does that mean?

OpenAI released solutions to three major math problems yesterday, one of which claims to refute the Exponential Time Hypothesis (ETH), a long-standing assumption in computer science about computational difficulty. The release drew a mix of excitement and concern from researchers. Anthropic mathematician Levent Alpöge suggested OpenAI's solutions were based on pre-existing work, and hours later Anthropic released its own refutation of ETH.
OpenAI structured the release differently for each of the three disciplines involved. For mathematicians, the work was posted on GitHub. Computer scientists formed a working group, requested the proofs be shared publicly and asked OpenAI to fund a conference to discuss them. Physicists formed a working group that will decide how their result is released.
MIT Professor Ryan Williams, a computer scientist focused on computational complexity, is organizing his field's response. Williams, who previously called an OpenAI proof dump in October a “DDOS attack on mathematicians,” agreed to collaborate with the frontier labs to help shepherd ideas vital to the field. He was asked by OpenAI to help interpret the solution alongside UC San Diego computer scientists Russell Impagliazzo and Ramamohan Paturi, who first published the ETH in 1999. Their first step was an open letter calling for feedback from interested scientists and discussion of norms around AI contributions.
Williams told TechCrunch that while some parts of OpenAI's proof might be called routine, two or three parts looked alien, and those parts convinced him the whole community needed to work on it together.
The ETH is part of the broader P vs NP question, which asks whether every problem whose solution can be quickly verified (NP) can also be quickly solved (P). A common example is Sudoku: a completed grid can be checked quickly, but finding the solution takes time. That gap underpins cryptography—with the key, a message can be decoded, but cracking a strong code is impractical. Most scientists assume P ≠ NP, and ETH provides evidence for that by positing that the fastest algorithm for a certain class of logic puzzles needs exponentially more time as variables increase.
OpenAI and Anthropic claim their models have found algorithms that solve those puzzles in sub-exponential time—faster than previously thought possible. If true, ETH would be refuted, potentially opening a path toward P = NP. Williams said every computer scientist should probably put a little more weight on the plausibility of P = NP, though that weight may vary. He added that the results call into question many intuitions about how tractable hard NP problems are, but he does not think OpenAI's algorithm could possibly be the best possible one.
The durability of ETH means many papers from the past decade may need revisiting if the claims hold up. Williams said this is partly why he pushed OpenAI to support a conference: the work will affect many researchers, and the community needs to understand how the foundations should change.
Before scientific concerns, there are cybersecurity questions. Williams said an improved algorithm could imply that cryptographic systems relying on certain key sizes are broken. He asked OpenAI science lead Dan Roberts whether any crypto systems were affected and whether anything is insecure now. According to Williams, OpenAI checked and was reasonably confident the result did not imply any breaks.
Williams remains optimistic despite concerns, especially about training the next generation of scientists. As a computer scientist doing high-level math, he may find the adjustment easier than pure mathematicians would. He said AI is forcing people to be more ambitious and to think about how to raise the bar for scientific questions when previous work becomes low-hanging fruit for AI.
This suggests a new paradigm for AI-driven scientific discovery is still taking shape, with norms around verification, credit and community review unresolved. In practice, developers and product teams should watch how the cryptographic implications are assessed, since any real break would have immediate security consequences. The conference and open letter signal that the computer science community intends to scrutinize these claims closely rather than accept them at face value.
Why it matters: If the refutations hold, they would challenge a foundational assumption in complexity theory and could eventually influence algorithm design and cryptography. For now, the immediate impact is on researchers who must evaluate the proofs, but developers relying on cryptographic systems should monitor follow-up analyses. The episode also highlights how AI labs are pushing into fundamental science, raising questions about how such results are validated and integrated into established fields.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
TechCrunch spoke to the MIT computer scientist charged with explaining OpenAI's latest mind-bending discovery.