A shift in OpenAI’s approach to releasing AI-generated research appears to be taking shape in its latest announcement, which presents different ways of involving external experts and academic researchers in interpreting findings and developing scientific understanding.
According to OpenAI, external experts recommended three distinct release strategies. These involved forming research groups to examine findings in computer science and physics, while for mathematics, the experts advised the company to release the AI-generated output publicly, allowing researchers to organise their own efforts to explain the result.
The announcement comes weeks after the Association for Human Mathematics (AHM) criticized OpenAI’s October 6 release of hundreds of AI-generated mathematical manuscripts and urged mathematicians to reject the work and “discontinue their work with OpenAI”. AHM argued that mathematicians had not asked for this work to be done.
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
https://t.co/7N6TPlft1P
— OpenAI (@OpenAI) October 6, 2026
AHM’s membership commitments call on members to refrain from working with commercial AI companies by providing technical labour, mathematical knowledge, consultation or publicity. Members also commit to not publishing mathematical texts generated by AI models, including papers, reviews, referee reports, lecture notes, teaching materials and books. Members of AHM’s AI-free caucus also commit to not using AI models in a research capacity.
“Our release process continues to evolve in response to feedback from the associated academic communities,” OpenAI said in its October 26 announcement.
OpenAI Claims Three New AI-Generated Discoveries in Math, Physics and Computer Science
The company presented three results spanning mathematics, computer science and physics, concerning harmonic analysis, the computational complexity of Boolean formula satisfiability, and quantum field theory, respectively.
We are announcing three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory.
https://t.co/9BX8HWQDTJ
— Dan Roberts (@danintheory) October 26, 2026
In mathematics, OpenAI said it had proved the local smoothing conjecture in every dimension, a result the company said also resolves the Kakeya and Stein restriction conjectures in all dimensions.
In computer science, OpenAI claimed its system had disproved the Exponential Time Hypothesis (ETH), a longstanding conjecture widely believed to be true. In physics, the company said the collaboration would explain a result that provides a mathematically rigorous existence proof for 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.
AI-Generated Manuscripts as ‘Raw Research Data
OpenAI described AI-generated manuscripts as “raw research data” that require further interpretation to become scientific understanding. As part of what it called a new approach, the company said it had reached out to external experts and asked them to act as “shepherds”, organising groups to understand the findings.
Different Strategies for Releasing Research
OpenAI said it was too early to determine which of the three release strategies would prove “most useful”. The company said it would provide the research materials and artifacts to the respective collaborations. While the computer science group decided to make its natural-language proof and Lean formalisation publicly available immediately through OpenAI’s GitHub repository, the physics group opted to withhold the AI-generated materials while preparing its own account of the findings.
For the mathematics result, OpenAI said it had decided to release the AI output publicly, following the same strategy it used for its October 6 release.
The company acknowledged that involving external experts could help interpret the findings and contextualise them within the broader academic community. However, it noted that “choosing initial shepherds necessarily shapes who gets involved first and creates asymmetry of information”.
Also Read: OpenAI Fires Three AI Safety Researchers Over “Breach of Trust”; METR Evaluation Under Scanner
