From the source
OpenAI says its internal models produced three research results in mathematics, computer science and physics, including a disproof of the Exponential Time Hypothesis.
- The company calls the AI-generated manuscripts raw research data that must still be converted into scientific understanding, and it is testing three release strategies.
- External experts acted as shepherds, advising that the mathematics output be published directly while the computer science and physics results get collaborator groups.
- The collaborations follow four principles: transparency with no nondisclosure agreements, no academic credit for OpenAI, recognition for each group's materials, and voluntary unpaid participation.
- OpenAI obtained the mathematics output on October 18, 2026, formalized it in Lean the next day, and published an AI writeup and Lean proof in its public
GitHub repository.
- The computer science disproof, produced on October 16, 2026 and formalized in Lean on October 17, will be shepherded by a group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams.
- OpenAI agreed to make all raw data public and to fund a conference on the work's implications, and says it does not yet know which release approach will prove most useful.