OpenAI announces three scientific discoveries and new academic collaboration process

On October 26, 2026, OpenAI announced three major AI-generated scientific discoveries across mathematics, theoretical computer science, and quantum field theory, including a refutation of the 1999 Exponential Time Hypothesis. To address past backlash over unverified findings and formalization errors in Lean, OpenAI introduced an academic shepherd process involving experts like David Gross and Ryan Williams. However, independent scientists and the AGMAI advisory panel express caution regarding early access access asymmetries.

OpenAI scientific discovery announcements

• ▪OpenAI announced that its internal AI models made three scientific discoveries across mathematics, computer science, and quantum field theory
• ▪OpenAI donated $5 million to the Simons Foundation to fund academic workshops focused on analyzing and understanding AI-produced scientific results.
• ▪OpenAI stated that its models produced a mathematically rigorous existence proof of interacting quantum field theories containing gauge fields and fermions in four dimensions, including Quantum Chromodynamics.

Academic verification skepticism

• ▪OpenAI withdrew three preprints and revised 14 others following a sign error discovered after its mass release of 722 preprints covering 372 math problems on October 6, 2026.
• ▪Researchers led by Anders Hansen at the University of Cambridge published a paper arguing that Lean-verified formalizations can compile successfully while failing to match the corresponding natural-language mathematical proofs.
• ▪The AGMAI panel of mathematical advisers recommended that AI companies refrain from granting selected scholars early access to AI-generated research outputs.

Shepherd-based publication process

• ▪OpenAI adopted an experimental review model using academic shepherds to shepherd and organize external research groups for analyzing raw AI outputs.
• ▪Computer scientists Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams issued an open letter stating their working group involvement does not constitute certification of OpenAI's Exponential Time Hypothesis claim.
• ▪A physics group led by Nobel laureate David Gross decided to defer releasing the raw AI physics data to write up their own academic interpretation.

Exponential Time Hypothesis refutation

• ▪OpenAI stated that its internal models disproved the Exponential Time Hypothesis, a theoretical computer science conjecture originally proposed in 1999 by Russell Impagliazzo and Ramamohan Paturi.
• ▪OpenAI released a paper describing a deterministic algorithm designed to count satisfying assignments of fixed-width CNF formulas in subexponential time, which would refute the Exponential Time Hypothesis in the 3-SAT case.

Mathematical conjecture proofs

• ▪OpenAI claimed its models proved Christopher Sogge's local smoothing conjecture for the Euclidean wave equation in every spatial dimension of at least two.
• ▪OpenAI reported that its proof of the local smoothing conjecture simultaneously resolves the Kakeya maximal and dimension conjectures and Stein's spherical restriction estimates across all dimensions.