OpenAI Shares AI‑Generated Proofs for Kakeya, ETH, and QCD

Introduction

OpenAI has unveiled three high‑profile research outputs that span pure mathematics, theoretical computer science, and theoretical physics. The releases consist of an AI‑generated proof of the local smoothing conjecture (and consequently the Kakeya conjecture), a formal refutation of the Exponential Time Hypothesis, and preliminary results concerning quantum chromodynamics and possible extensions of the Standard Model. Rather than presenting polished papers, OpenAI treats these artifacts as raw research data and is experimenting with collaborative pathways that let external experts shepherd the interpretation, verification, and dissemination of the findings. The approach is designed to balance openness with the need for expert guidance, while explicitly stating that OpenAI does not claim academic credit for the discoveries.

Mathematical Breakthrough: Local Smoothing and the Kakeya Conjecture

The mathematics component addresses a long‑standing problem first articulated by Charles Sogge in 1991. The local smoothing conjecture describes how wave energy behaves when it is focused, asserting that extreme focusing events are short‑lived and that average focusing is far milder than worst‑case behavior. Proving this conjecture automatically resolves related conjectures such as the Kakeya problem and Stein’s restriction theorem across all spatial dimensions. In 2019, Guth, Wang, and Zhang succeeded in two dimensions, and in 2025 Wang and Zahl extended the Kakeya result to three dimensions. OpenAI’s models have now produced a proof that works in every dimension, leveraging a novel auxiliary partial differential equation. The AI output was generated on 18 October 2026 and formalized in the proof assistant Lean the following day. Following the recommendation of an external expert, OpenAI chose a “release‑the‑output” strategy, publishing both a natural‑language exposition and the Lean formalization directly on its public GitHub repository. This makes the raw proof immediately accessible for independent scrutiny, replication, and further development by the mathematical community.

Computer Science Impact: Refuting the Exponential Time Hypothesis

The computer‑science release tackles the Exponential Time Hypothesis (ETH), a conjecture that has underpinned hundreds of conditional hardness results in graph theory, computational biology, databases, and geometry. ETH posits that Boolean formula satisfiability cannot be solved in sub‑exponential time, a assumption closely tied to the broader P versus NP question. OpenAI’s system has produced a proof that ETH is false, a result that has been verified using the Lean proof assistant. By invalidating ETH, the work reopens the possibility that many supposedly hard problems admit faster algorithms than previously imagined. To support community investigation, the collaboration has elected to make the natural‑language proof and its Lean formalization publicly available on GitHub, while also forming an external group of collaborators to explain the implications and guide further research. This dual approach ensures that both the raw AI output and expert interpretation are accessible, encouraging rapid iteration and validation.

Physics Insight: Quantum Chromodynamics and Standard Model Completions

The physics component focuses on the existence of quantum chromodynamics (QCD) and potential completions of the Standard Model. OpenAI generated AI‑produced artifacts that illuminate these topics, but the organization has opted for a more guarded release strategy. An external group of collaborators has been assembled, and they are currently deciding when‑if ever‑to publish the AI‑derived materials. The emphasis is on allowing domain experts to shape the narrative, ensure proper contextualization, and manage the timing of disclosure in a way that maximizes scientific benefit while respecting the sensitivity of the underlying theoretical work.

Collaboration Model and Best Practices for the Community

OpenAI’s three‑pronged release framework illustrates a pragmatic response to the challenges of integrating machine‑generated research into established scientific pipelines. The core principles are:

• Transparency – No non‑disclosure agreements; participants may discuss, share intermediate results, and publish conclusions freely.
• No Academic Credit for OpenAI – The organization explicitly disclaims ownership of the discoveries, ensuring that credit flows to the researchers who interpret and extend the work.
• Recognition for Collaborators – The groups that shepherd the results receive acknowledgment for their write‑ups, analyses, and any downstream publications.
• Independence – Involvement is voluntary and unpaid; contributors speak for themselves, and OpenAI does not act as their spokesperson.

For IT, security, and engineering professionals, this model suggests several actionable steps. First, monitor the GitHub repositories where the AI outputs are deposited; the Lean formalizations provide a concrete verification path that can be integrated into automated testing pipelines. Second, consider joining or forming external collaboration groups for the computer‑science and physics releases, as participation offers early access to cutting‑edge theoretical advances that may influence algorithm design, cryptographic assumptions, and high‑performance computing simulations. Third, adopt a culture of open verification: treat AI‑generated proofs as raw data, apply formal methods where possible, and contribute improvements back to the community. Finally, provide feedback to OpenAI on the effectiveness of each release strategy, helping to refine how future machine‑driven discoveries are shared.

Conclusion

By exposing AI‑generated mathematical, computational, and physical insights to community scrutiny, OpenAI is creating a living laboratory for scientific validation that aligns well with modern DevSecOps practices. Leveraging such open, collaborative frameworks can help organizations turn cutting‑edge AI research into reliable, production‑ready solutions.