OpenAI announced three purported scientific discoveries in mathematics, computer science, and physics, alongside an evolving release process designed to engage external academic experts. The mathematical result concerns harmonic analysis, where the company claims its models produced a proof of the local smoothing conjecture in every dimension, thereby resolving the Kakeya conjecture and Stein’s restriction conjecture in all dimensions. The computer science breakthrough involves a natural language disproof for the Exponential Time Hypothesis (ETH), which the company stated was formalized in Lean on October 17th, 2026. An independent working group of computer scientists, including Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams, will examine this result, and at the group’s request, OpenAI is making the materials public immediately on GitHub. For the physics result, the company stated it achieved a "mathematically rigorous existence proof of certain interacting quantum field theories containing gauge fields and fermions in four dimensions, including Quantum Chromodynamics (QCD) and a unified completion of the Standard Model of particle physics," which a group led by David Gross, including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov, will shepherd.
OpenAI detailed three distinct release strategies for these findings, adapting its process based on feedback from academic communities. For the mathematics result, the company advised releasing the AI output for researchers to organize their own efforts to explain the result. For the computer science result, the plan involves forming an external group of collaborators to explain the findings while making the AI output publicly available. For the physics result, the company plans to form an external group and defer to them on the release of materials.
The company outlined four principles governing these external collaborations: Transparency, ensuring participants are free to discuss the work and publish conclusions without signing nondisclosure agreements; No Academic Credit, stating that OpenAI does not seek academic credit for discoveries made by its internal models; Recognition, which mandates that the collaborating group receives credit for the understanding they develop and materials they author; and Independence, confirming that participation is voluntary and unpaid, with members acting independently.
Regarding the release mechanics, the group for the computer science collaboration decided to make the natural language proof and Lean formalization public immediately, publishing them directly to GitHub. Conversely, for the physics collaboration, the group decided to withhold releasing the AI-produced artifacts and instead focus on writing up the material themselves. OpenAI noted that these approaches involve trade-offs, as empowering experts can shape who gets involved first, creating an asymmetry of information.
In addition to the technical breakthroughs, OpenAI released an open letter seeking community input on broader questions, including what expectations should govern early access to AI-generated results, how credit should recognize both new discoveries and the work of checking or simplifying them, and how to support researchers as methods change.
