OpenAI claims it shattered a 30-year math assumption. Experts aren't sold.

OpenAI dropped a bombshell Monday: its models found that a commonly held computer science hypothesis, the Exponential Time Hypothesis (ETH), is wrong. Hours later, rival lab Anthropic released its own research refuting ETH. The catch: the independent algorithm experts OpenAI asked to organize an open review have not certified the results, and they stress their involvement is not an endorsement.

That matters because ETH is the nearly 30-year-old belief that certain logic problems become overwhelmingly time-consuming as they grow. If OpenAI's announcement holds, it would likely reopen the hunt for increasingly capable algorithms, something computer scientists had largely assumed couldn't get much better. In plain English, it could mean that many problems once considered too hard to solve quickly might actually be solvable with smarter methods.

For context, ETH sits on the same turf as P versus NP, the most famous open question in computer science and one of the $1 million Millennium Prize Problems. If P equals NP, problems that currently take forever to compute could be solved efficiently, with massive implications for cryptography, logistics, drug discovery, and AI itself. OpenAI's claim doesn't resolve P = NP, but as MIT professor Ryan Williams put it, if true, it should make every computer scientist put at least a little more weight on the possibility of P = NP.

Williams, one of the three independent algorithm design experts researching OpenAI's claims, has not certified the results. He wrote on X that, in practical terms, the claimed breakthrough could lead to "faster algorithms." He has previously said that he doesn't believe in ETH, which makes his cautious interest notable.

The rest of the internet's nerdiest corners are split. Gary Marcus, an emeritus professor at NYU and a frequent AI skeptic, wrote on X that this could be "a big deal if correct." John D. Cook, a Ph.D. and tech consultant in applied mathematics, was more measured in a blog post: "The status of P = NP is still unknown. While 3-SAT does not take exponential time, it still takes more than polynomial time as far as anyone knows." Cook's point is that even if ETH falls, there is a gap between exponential and polynomial time that remains unresolved.

Some in the field are openly impressed. Joshua Fonseca Rivera, an AI Safety Research Fellow at Anthropic, wrote on X that the claim was a reminder that humans could have underestimated a computer's intelligence: "The bitter lesson should make this unsurprising. I know this logically, but... wow, we really don't know anything." Fields Medal-winning mathematician Timothy Gowers was blunt: "Very surprising and consequential that it is false." Dan Shipper, CEO at Every, captured the general mood: "I regret not getting a Ph.D. in frontier math. (Not that I ever considered getting one but still)."

For executives, the immediate stakes are less about whether ETH is dead and more about how AI labs are now setting the research agenda for theoretical computer science. OpenAI and Anthropic, two leading AI labs with massive compute resources, are using AI models to make conjectures about math problems that humans have struggled with for decades. Even if these specific claims fall apart under review, the pattern matters: AI is increasingly being used as a research engine, not just a content generator. That shift could accelerate progress in algorithm design, optimization, and possibly cryptography, which would ripple across every industry that depends on secure data transmission or complex supply chains.

The review process is the next thing to watch. OpenAI said it had asked a group of outside computer scientists to organize an open review of its claim, and the researchers have highlighted that their involvement is not an endorsement. That is a healthy sign of rigor, but it also means the verdict is genuinely open. If the AI labs' announcements hold, the hunt for increasingly capable algorithms would reopen, and the payoff could be enormous. If they do not, the episode will still stand as a reminder that AI is now generating hypotheses at the frontier of human knowledge, and that the smartest people in the room are increasingly likely to be machines.