In brief
- OpenAI is seeking external experts’ evaluations of AI-generated proofs in three different fields.
- The computer science paper argues that the Exponential Time Hypothesis is false; the claim will be independently reviewed.
- The mathematical proof aims to show that the local smoothing conjecture holds in every dimension.
- The physics paper focuses on the mathematical existence of Yang-Mills theory; it does not claim to solve the mass gap problem.
On October 27, 2026, OpenAI announced that it had submitted three proofs that it says its AI models generated in mathematics, computer science and physics for evaluation by external experts. The company is treating these works not as completed scientific results, but as research material that needs to be reviewed. The evaluation process differs across the three fields: some proofs have been published, while the materials for the physics work have not been released for now.
What is the claim in computer science?
The first paper examines the Exponential Time Hypothesis (ETH), which is closely related to the Boolean satisfiability problem (SAT). ETH posits that, in the worst case, the time required to solve SAT grows exponentially with the number of variables. OpenAI says its models produced a proof showing that this hypothesis is false. If confirmed, the claim could prompt a reassessment of the foundations of some research that establishes bounds by assuming ETH; however, it would not automatically invalidate that research.
Russell Impagliazzo and Ramamohan Paturi of UC San Diego, along with Ryan Williams of MIT, will work to independently review and explain the result. At the group’s request, OpenAI published the written proof and the proof in Lean format. Lean can check whether a proof constructs the encoded statement step by step, but it cannot, on its own, determine whether that statement accurately represents the problem the researchers set out to solve. Expert review therefore involves more than formal verification: the scientific meaning of the claim must also be assessed.
What results stand out in mathematics and physics?
The mathematics paper concerns the local smoothing conjecture, which deals with how waves propagate over time. OpenAI says its models produced a proof of the conjecture that covers every dimension, not just two. According to the company, this result could also establish the Kakeya conjecture and Stein’s restriction conjecture in all dimensions. No dedicated expert group has been formed for this work; the written proof and Lean proof have been made available for researchers to review.
The physics paper focuses on the mathematical foundations of quantum field theories. OpenAI says its models proved the mathematical existence of certain theories in four dimensions. The result put forward for Yang-Mills theory covers only the existence part of the Millennium Problem. It does not claim to prove that the theory has a mass gap, so it does not mean the Millennium Problem in question has been solved in full. An expert group led by David Gross will evaluate the work; the company is not releasing the physics proof’s materials for now. The group is expected to produce the result through its own work.
What this means for architecture, interior design and visualization workflows
This development is not an announcement of a new feature added to architectural modeling or visualization software. The validity of the proofs and their practical implications are still subject to expert review. OpenAI says the computer science result could help research into better algorithms, but acknowledges that more work is needed before it can lead to practical gains. As a result, no changes to specific workflows, hardware requirements or software compatibility have been announced for architecture firms and visualization teams at this time.
OpenAI also announced that it is allocating $5 million to the Simons Foundation to support discussion of the implications of AI-generated results. The company argues that scientific credit should go to the research groups that explain and evaluate the proofs. How the process will be verified, who will access the results and when, and how scientific contributions should be defined remain subjects of debate.
Sources
1 sourceSource texts are not republished; short quotes are marked, everything else is our own summary and commentary.
For architecture and visualization teams, this announcement does not mean an immediate change to production tools or workflows. The key issue is how to design a verification process for technical results generated by AI. That process could eventually contribute to the development of computational methods, but, as the source also notes, practical benefits remain uncertain.
For firms in Turkey, there is currently no basis for making new hardware investments, changing licenses or taking software compatibility decisions. It makes more sense to follow the independent review of the claims, particularly the distinction between theory and practical outcomes. Their implications should not be treated as settled until experts publish their assessments.
Frequently asked questions
In which fields did OpenAI’s AI produce proofs?
OpenAI announced three works, in mathematics, computer science and physics. Independent experts are evaluating the validity of the proofs.
Is OpenAI saying it has solved the Millennium Problem?
No. The Yang-Mills work is said to cover only the mathematical existence of the theory. The mass gap has not been proved.
Does a Lean proof, by itself, prove that a result is correct?
No. Lean checks the proof steps for the mathematical statement as written; experts must assess whether that statement accurately represents the intended problem.
