OpenAI and Anthropic say AI models challenged the Exponential Time Hypothesis. Experts are reviewing the proof and its implications.
OpenAI and Anthropic have presented AI-generated work that challenges the Exponential Time Hypothesis, an assumption in computer science. The hypothesis says certain difficult problems require exponentially more time as their size grows. OpenAI says its model produced a refutation, while independent researchers are still testing the result. The outcome could affect complexity theory, algorithms and the way scientists verify AI discoveries.
What the AI claim means for computing
The Exponential Time Hypothesis, or ETH, concerns the difficulty of solving Boolean satisfiability problems. In simple terms, these problems ask whether a set of logical statements can be made true at the same time. Researchers have long believed that some versions cannot be solved substantially faster than exponential-time methods in the worst case.
OpenAI says its model found an algorithm running in sub-exponential time for the relevant problem class. If the argument survives review, it would overturn ETH, which has supported conditional results in graph theory, computational biology, databases and computational geometry. 
OpenAI’s announcement about the ETH result. The claim remains as a settled scientific result because review is still underway.
ETH is related to P versus NP
The development connects to the famous P versus NP problem, but it does not solve it. P versus NP asks whether every problem whose answer can be checked quickly can also be solved quickly. ETH is a narrower assumption about the time needed for particular hard problems, so refuting it would not automatically prove that P equals NP.
That distinction matters for the public discussion. A theoretical improvement may still leave an algorithm too slow for everyday use. It also does not mean that online banking or common encryption systems have suddenly failed. TechCrunch reported that MIT computer scientist Ryan Williams said OpenAI had checked for possible effects on cryptography and was reasonably confident the result did not imply immediate breaks.
Researchers are examining the proof
OpenAI has released a natural-language write-up and a formal proof in Lean, a system that checks whether each step follows logically from the stated rules. Formal verification adds an important layer of confidence, but it does not replace independent examination of the argument’s assumptions, interpretation and wider consequences. 
OpenAI’s public mathematics and proof repository
A review group led by Russell Impagliazzo and Ramamohan Paturi of the University of California, San Diego, and Williams of MIT will study the computer-science result. Impagliazzo and Paturi helped propose ETH with Francis Zane in 1999. OpenAI says the collaboration will be transparent, voluntary and independent, with no nondisclosure agreements or academic credit claimed by the company.
“Every computer scientist should probably put a little more weight on the plausibility of P = NP, but one’s weight may vary.” — Ryan Williams, MIT, as quoted by TechCrunch
AI changes the research process
The episode highlights a growing challenge for scientific publishing. AI systems may produce technical results faster than humans can explain, check or place them in context. That creates questions about credit, access to raw materials and who should be responsible when a model’s reasoning contains a hidden mistake.
The Next Web reported that the review group asked OpenAI to publish raw data and support a conference about the result. OpenAI agreed to those requests and said it would not seek academic credit for the discovery. The company also said its models produced other results in mathematics and physics, though those claims are following separate review processes.
For broader technology coverage, readers can also see Diaspora Digital Media’s report, 
“Bank of America Warns Customers Not to Share Banking Passwords With AI Assistants”, which examines a different risk created by rapidly advancing AI systems.
What happens next
The immediate task is not to redesign software or abandon current security systems. Researchers must determine whether the proof is correct and what consequences follow for earlier papers. Even a valid refutation may produce theoretical insight before practical improvements.
Ultimately, the announcement is significant because it tests both a central idea in computer science and a new model for AI-assisted research. OpenAI has made its materials, while experts are asking the wider community to inspect them. The result will matter most if independent researchers can reproduce its logic and map its consequences. Until then, the appropriate description is a remarkable claim under review.
