On October 16, 2026, OpenAI says one of its systems produced a natural-language disproof of the Exponential Time Hypothesis, or ETH, a long-standing assumption about the difficulty of solving certain computational problems. The company says it formalized the result in Lean the following day, then moved to organize outside review rather than declare a settled scientific victory.¹

ETH has shaped hundreds of conditional hardness results across computer science. If the claim holds, it could force researchers to revisit assumptions in fields from graph theory to computational biology—and could reopen broader questions around P versus NP. OpenAI says the finding may guide the search for faster algorithms, while stressing that practical consequences remain uncertain and require “significant further research.”¹

The release immediately exposed the tension in AI-assisted science: breathtaking output, but no shortcut around verification. OpenAI released the proof materials publicly and agreed to fund a conference after a group led by MIT’s Ryan Williams and UC San Diego’s Russell Impagliazzo and Ramamohan Paturi asked for an open examination. The company calls the manuscripts “raw research data,” not finished scientific understanding.¹

Williams, who is helping organize the response, said parts of the proof looked routine—but “two or three alien-looking parts” convinced him that “the whole community needs to work on [it] together.”² He has also warned that a credible refutation could unsettle work built on ETH and prompt questions about whether some cryptographic systems are affected, though OpenAI told him it had found no immediate security break.

The episode arrives as frontier labs face pressure on more than their science. Anthropic’s expected IPO is being watched as a referendum on the durability of the AI boom, after a broader sell-off and concerns over enormous infrastructure spending.³ For OpenAI, the ETH announcement offers a striking counterpoint: the promise that AI can generate foundational knowledge. For academia, it is a test of whether that promise can be made legible, reproducible and trusted.