OpenAI Claims to Have Broken the Exponential Time Hypothesis

Computer scientists have spent decades treating the Exponential Time Hypothesis, or ETH, as an immovable brick wall. It posits that specific computational problems are inherently too difficult for any algorithm to solve quickly, regardless of hardware speed or coding elegance. This assumption has defined the limits of what we thought was possible. Now, OpenAI is claiming that wall is an illusion. According to their latest announcement, their AI models have generated a proof demonstrating that the hypothesis is actually false.

This is not a minor technical fix or a niche update. The Exponential Time Hypothesis sits under the foundation of hundreds of research papers, spanning database optimization, computational biology, and geometry. If the AI is correct, the entire framework for measuring difficulty in these fields is flawed. We may have been optimizing for the wrong constraints for years. The company argues this result, paired with breakthroughs in pure mathematics and quantum physics, signals a shift where AI moves from assisting research to generating raw, verified scientific discovery.

A Shock to the Foundations of Complexity

The central claim in computer science is that the Exponential Time Hypothesis is false. This conjecture has long served as the benchmark for computational possibility, asserting that Boolean formula satisfiability requires exponential time. Researchers rely on this assumption to derive conditional hardness results, proving that certain shortcuts are impossible. OpenAI now suggests those shortcuts may exist, waiting to be found.

The implications are vast. If the hypothesis is false, it implies we could theoretically find better algorithms to solve complex problems more efficiently. This would revolutionize fields that currently hit a dead end when scaling up. The company notes that the scientists who originally proposed the hypothesis are now part of the group tasked with examining the AI-generated output. It is a surreal moment for the academic community, where the originators of a theory are now checking the work of a machine that claims to have broken it.

Beyond the Code

The announcement extends beyond computer science. OpenAI claims its models have also produced a proof for the local smoothing conjecture in every spatial dimension. This is a deep problem in harmonic analysis that has challenged mathematicians for decades. The conjecture, proposed by Sogge in 1991, is linked to the Kakeya conjecture and Stein's restriction conjecture. The AI generated a proof on October 18, 2026, and it was formalized in the Lean proof checking system the very next day.

The math result is just as significant. The proof is based on a novel auxiliary partial differential equation. OpenAI has published both the natural language writeup and the formal Lean proof on its public GitHub repository. This transparency is a deliberate choice. They want researchers to look at the raw data and make their own efforts to explain it. It is a shift from asking the AI for answers to inviting the global scientific community to audit the process.

Physics and the Millennium Prize

The third pillar of this announcement dives into theoretical physics. OpenAI says its models developed existence proofs for certain four-dimensional quantum field theories. This includes quantum chromodynamics, which describes the strong nuclear force holding quarks together. The work also includes a proposed proof of the existence of pure Yang-Mills theory. This addresses a part of one of the Clay Mathematics Institute's Millennium Prize problems, a challenge that has remained unsolved for over twenty years.

Nobel Prize-winning physicist David Gross is leading the group of researchers examining these physics findings. Unlike the math and computer science results, the physics materials have not been publicly released yet. The company deferred the release to the group handling the verification. They have provided an email address for questions about their efforts. This cautious approach suggests that even the most confident AI claims require human eyes before they can be considered settled science.

A New Model for Scientific Verification

OpenAI is experimenting with a new approach to how these discoveries are handled. They describe the AI-generated manuscripts as raw research data. Their goal is to convert this data into scientific understanding through collaboration. For the computer science results, they asked the external expert group to explain them in their own writeups and materials while also releasing the AI output to the public. This dual track allows for both immediate access and deep expert analysis.

The company has pledged financial support for this verification process. They are welcoming feedback on how to broaden participation and make the process more useful. This is a departure from the typical closed-door model of big tech research. By opening the floor to independent academic groups, OpenAI is trying to build trust. They are not seeking academic credit for the models' discoveries. Instead, the credit goes to the people in each group who create the writeups and materials. It is a bold move to integrate AI into the human workflow of science.

The Road Ahead

This announcement comes just weeks after OpenAI released manuscripts with over 700 pages of solutions for 372 major math problems. The trajectory is clear. The company is pushing the boundaries of what AI can do in abstract reasoning and complex problem solving. The reaction from the scientific community will be the true test. If the proofs hold up, it will be a landmark moment in the history of science. If they fall apart, it will be a cautionary tale about the limits of machine intelligence.

For now, the ball is in the court of the independent researchers. The raw data is out there, and the experts are working through it. The Exponential Time Hypothesis has been a cornerstone of computer science. Disproving it would rewrite the textbooks. We are watching a potential paradigm shift unfold in real time. The next few months will determine whether this is a genuine breakthrough or a spectacular hallucination. Either way, the conversation has changed forever.