OpenAI dropped a bombshell on October 26, 2026, announcing three purported discoveries in mathematics, computer science, and physics. The centerpiece of the announcement is a claimed proof that the Exponential Time Hypothesis (ETH) is false, a foundational conjecture in complexity theory that has guided algorithmic research for decades. The company stated that its models also proved the local smoothing conjecture in every dimension and identified the existence of certain interacting quantum field theories in four dimensions.

The Complexity Theory Shockwave

The refutation of the ETH is arguably the most disruptive claim for the computer science community. According to reports from The Verge, the resulting paper describes a deterministic algorithm that exactly counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space, specifically refuting the hypothesis in the 3-SAT case. This is not a minor tweak to existing bounds; it challenges the conditional hardness results used in hundreds of papers. OpenAI has released the raw AI output for this work on GitHub, but the formal verification and explanation are being shepherded by an external group led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams. Notably, Impagliazzo and Paturi are the original proposers of the hypothesis, lending significant weight to the scrutiny that will follow.

Harmonic Analysis and the Lean Proof

In the realm of pure mathematics, OpenAI claims to have proven the local smoothing conjecture in every dimension, with implications for Kakeya and Stein problems. The timeline for this result shows OpenAI obtained the AI output on October 18, 2026, and successfully formalized the proof in the Lean theorem prover the following day. This approach differs from the ETH work; for the math result, the raw AI output was released immediately, allowing researchers to organize their own efforts to understand the proof without an intermediary 'shepherd' group controlling the narrative. This transparency strategy aligns with OpenAI’s stated principles of no academic credit for the company and independence for the external reviewers.

Physics on Hold and Expert Skepticism

The third discovery, concerning quantum field theory, is being handled differently. The physics group, led by David Gross, has decided to hold off on releasing the AI-produced artifacts, deferring to their own judgment on when and how to share the materials. This cautious approach comes amid growing scrutiny from the academic community. The AGMAI panel of math advisers recently recommended that AI companies avoid designating specific individuals to receive output first, citing concerns about unequal access and borrowed reputation. OpenAI acknowledged this feedback, noting that it does not yet know which of its three distinct release strategies will prove most effective for scientific validation.

Key Takeaways

  • OpenAI claims to have disproved the Exponential Time Hypothesis (ETH) using a deterministic subexponential algorithm for 3-SAT.
  • The ETH refutation is being verified by a group led by the hypothesis's original proposers, Impagliazzo and Paturi.
  • A proof for the local smoothing conjecture in all dimensions was formalized in Lean within 24 hours of AI generation.
  • The physics discovery regarding 4D quantum field theories remains unreleased, with the expert group controlling the disclosure timeline.
  • OpenAI is funding a conference on the implications of these findings and has released raw data on GitHub for public scrutiny.

The Bottom Line

If this ETH refutation holds up to the brutal scrutiny of Impagliazzo and Williams, it’s not just a benchmark winβ€”it’s a fundamental rewrite of complexity theory's rulebook. We are no longer talking about AI solving Sudoku; we are talking about AI potentially breaking the assumed limits of computational hardness.