OpenAI Models Claim Disproof of Exponential Time Hypothesis Alongside Math and Physics Breakthroughs

On October 16, 2026, an internal artificial intelligence model at OpenAI generated a natural-language document claiming to dismantle the Exponential Time Hypothesis. The following day, the company ran the machine's logic through Lean, a rigorous computational theorem prover, which verified the proof's structural integrity. The result targets a foundational assumption in computer science that has stood unchallenged for decades.[1][3][4]

OpenAI publicly announced the result on October 26, alongside two other massive claims in pure mathematics and theoretical physics. Rather than formatting the discoveries as traditional academic papers, the company uploaded the machine-generated code and text to GitHub. OpenAI described the outputs as "raw research data," testing whether human scientists can extract meaning from alien, algorithmic logic.[1][2][6]

The Exponential Time Hypothesis (ETH) was first proposed in the late 1990s by Russell Impagliazzo, Ramamohan Paturi, and Francis Zane. It essentially argues that certain complex optimization problems—specifically Boolean satisfiability—cannot be solved significantly faster than checking every possible answer in exponential time. This assumption forms the bedrock of modern computational complexity theory.[1][3]

Over the last thirty years, researchers have published hundreds of papers relying on ETH to prove that specific problems in computational biology, graph theory, and database architecture are fundamentally hard to compute. If OpenAI's disproof is accurate, it implies that faster, highly efficient algorithms exist for these bottlenecks. This could eventually accelerate software performance across multiple data-heavy industries.[1][3][7]

Independent scrutiny for algorithmic claims

To evaluate the claim, OpenAI recruited an independent group of experts, including Impagliazzo and Paturi themselves, alongside MIT's Ryan Williams. The researchers emphasized that their involvement does not certify the AI's work. "A result of this significance requires careful, independent scrutiny," the three scientists wrote in an open letter to the computer science community.[3]

The review group is operating without non-disclosure agreements and receiving no payment from the company. They have secured OpenAI's commitment to release all raw data associated with the ETH result and to fund a community conference to dissect the findings. The goal is to translate the machine's brute-force verification into human-readable scientific understanding.[3][4]

Beyond computer science, the AI models tackled harmonic analysis. OpenAI claims the system proved Christopher Sogge's 1991 local smoothing conjecture for wave equations in every spatial dimension. Human mathematicians had previously managed to prove the conjecture only in two dimensions, achieving that milestone in 2019 after decades of incremental progress.[1][3]

Because mathematical conjectures are deeply interconnected, the local smoothing proof acts as a skeleton key. According to the release, the AI's solution simultaneously resolves the Kakeya conjecture and Stein's restriction conjecture in all dimensions. Prior to this, the Kakeya conjecture had only been proven up to three dimensions, a breakthrough reached by human researchers in 2025.[1][3]

Tackling the Standard Model

The third pillar of the announcement involves theoretical physics. OpenAI stated that its models generated a natural-language proof for the existence of Yang-Mills theory on September 9, 2026. Nine days later, on September 18, the system subsequently produced a proof for Quantum Chromodynamics (QCD), aiming to provide a mathematically rigorous foundation for the Standard Model of particle physics.[1][5]

The company delayed the physics announcement for over a month to allow the Lean software to formalize the Yang-Mills existence proof. That computational verification was completed on October 16. The physics claim is arguably the most ambitious of the three, attempting to bridge the gap between abstract quantum field theories and rigorous mathematical existence in four dimensions.[1][4]

OpenAI is deploying three distinct strategies to handle the sheer density of the machine's output. For the mathematical conjectures, the company simply dumped the AI output online, leaving the global math community to organize its own review. For the computer science claim, it released the data while formally appointing the Impagliazzo-led group to shepherd the analysis.[1][4][7]

The physics release is the most guarded. OpenAI formed a collaboration group led by Nobel laureate David Gross, alongside physicists Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov. However, the company deferred entirely to this group on when and how to release the materials, keeping the raw physics data unpublished while the experts begin their review.[1][7]

Data without understanding

While the corporate framing suggests a historic triumph, the reality of the release highlights the limitations of generative models. OpenAI explicitly stated it seeks no academic credit for the discoveries, acknowledging that generating a verified proof in Lean is not the same as producing scientific understanding. The AI has provided the answers, but the reasoning remains locked in code.[1][4]

OpenAI outlined four principles for these collaborations: transparency, no academic credit for the company, recognition for the human reviewers, and strict independence. The company noted that inviting a researcher to steward a collaboration does not imply that the researcher made the original discovery, nor does it mean OpenAI speaks for the scientists.[4]

The announcements highlight a growing tension in theoretical research. If brute-force computational scaling can resolve century-old conjectures that stumped generations of human experts, the bottleneck in scientific discovery may no longer be human ingenuity, but compute power. The academic community must now decide how to integrate machine-generated truths that no single human fully comprehends.[2][7]

The immediate focus shifts to the independent review groups. Over the coming months, mathematicians and physicists will attempt to reverse-engineer the AI's proofs, searching for novel techniques or hidden flaws. If the Exponential Time Hypothesis is definitively shattered, the race will begin to find the hyper-efficient algorithms the AI's proof suggests must exist.[3][7]