OpenAI Challenges Decades of Computer Science by Refuting Foundational Complexity Rule
San Francisco, Tuesday, 27 October 2026.
OpenAI announced its artificial intelligence disproved the longstanding Exponential Time Hypothesis, publishing machine-verified proofs that suggest complex computational problems may be solved significantly faster than previously assumed possible.
A Landmark Claim with Far-Reaching Implications
On October 20, 2026, OpenAI (private, no ticker symbol) asserted that an internal artificial intelligence model had produced a machine-verified refutation of the Exponential Time Hypothesis (ETH), a foundational conjecture in computational complexity theory proposed in 1999 by Russell Impagliazzo and Ramamohan Paturi of UC San Diego [alert! ‘The ETH was proposed in 1999 by Impagliazzo and Paturi, not in 1991 as some conjectures were’] [5][7]. If validated, the claim would imply that certain NP-hard Boolean satisfiability problems—including 3-SAT—can be solved in subexponential time, challenging decades of theoretical assumptions about computational limits [13][14]. The ETH posits that solving k-SAT requires time exponential in the number of variables, a bedrock assumption underpinning algorithmic efficiency research [7][14]. OpenAI’s formalization of the disproof was machine-verified using the Lean theorem prover on October 17, 2026, one day after the natural-language proof was generated on October 16, 2026 [1][5][7].
Verification and Independent Scrutiny
OpenAI did not claim academic credit for the discovery and instead released the natural language proof and Lean formalization publicly on GitHub on October 25, 2026, inviting independent review [1][5][7]. An external collaboration led by Impagliazzo, Paturi, and Ryan Williams of MIT was formed to conduct open, unpaid, and independent scrutiny of the claim, with the researchers emphasizing that their involvement was not a certification of correctness [1][2][7][14]. ‘Our involvement should not be taken as certification of the claim. A result of this significance requires careful, independent scrutiny,’ the group stated in a public post [1][2]. The Lean formalization was released alongside the natural language manuscript, enabling global verification by the formal methods community [5][7]. OpenAI’s approach aligns with its stated principles of transparency and independence in releasing AI-generated scientific results [7].
Broader Scientific Context: Mathematics and Physics Breakthroughs
The ETH refutation is part of a triad of AI-generated scientific advances announced by OpenAI on October 26, 2026, which also includes a proof of the local smoothing conjecture in all dimensions (resolving the Kakeya and Stein restriction conjectures) and existence proofs for quantum chromodynamics (QCD) and pure Yang–Mills theory [1][2][4][5][6][7][14]. The local smoothing result, formalized in Lean on October 19, 2026, builds on prior work by Larry Guth, Hong Wang, and Ruixiang Zhang, who proved the conjecture in two spatial dimensions in 2019, and Wang and Zahl, who resolved the Kakeya conjecture in three dimensions in 2025 [5][7][14]. OpenAI’s model generated the natural-language proof for the local smoothing conjecture on October 18, 2026, one day after completing a Lean formalization for the Yang–Mills existence proof, first generated on September 9, 2026 [5][7][14]. The physics results, including QCD and a completion of the Standard Model, are being independently written up by a collaboration led by Nobel laureate David Gross and includes Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov [1][2][5][6][7].
OpenAI’s Release Strategy and Broader Impact
OpenAI’s release strategy for mathematical results emphasizes transparency and community verification. The organization published AI-generated papers and Lean proofs for the ETH refutation and local smoothing conjecture on its public GitHub repository, allowing researchers worldwide to examine, replicate, and build upon the findings [1][5][7]. In contrast, the physics results are being withheld until the external collaboration completes its peer review and write-up [1][2][5][6][7]. The company also announced a $5 million donation to the Simons Foundation to fund workshops and conferences focused on AI-generated mathematics and physics results, supporting independent scrutiny and community collaboration [5][7]. OpenAI framed the manuscripts as ‘raw research data,’ emphasizing their provisional nature and the need for further verification [12]. The announcement follows the release of 722 AI-generated math papers on October 6, 2026, and comes amid growing calls from the academic community for better citation practices and citable repositories for AI-generated results [5][7].