OpenAI's AI Claims to Disprove ETH and Solve Kakeya Conjecture, Sparking Academic Debate

Key points

• OpenAI claims its AI models have disproved the Exponential Time Hypothesis (ETH), a foundational assumption in computational complexity theory regarding the time required to solve 3-SAT problems.
• The company asserts it has resolved the Kakeya conjecture by proving the local smoothing conjecture for all dimensions, a major problem in harmonic analysis and geometry.
• OpenAI also claims to have established existence results for certain interacting four-dimensional quantum field theories, including Quantum Chromodynamics (QCD) and a unified completion of the Standard Model.
• The company is releasing model-generated texts as 'raw research data' rather than seeking academic credit, asking external experts to act as 'shepherds' to convert these outputs into human knowledge.
• The physics group is withholding AI-produced artifacts to write up the results themselves, while math results have been formalized in the Lean proof assistant, though some discrepancies in verification scope have been noted.

Background

This announcement follows a turbulent period for AI in science, including OpenAI's retraction of three math preprints in early October 2026 after academic backlash over the use of unreleased internal models. The current release addresses previous concerns by explicitly stating OpenAI does not seek academic credit for the discoveries and is working with external experts to verify the results.

How outlets are covering it

The Verge's coverage was dominated by an unrelated advertisement for SKF's 'Pay Like It's 1919' campaign, offering no substantive analysis of the OpenAI announcement. Gizmodo and Kingy AI provided detailed technical breakdowns. Gizmodo highlighted the potential implications of disproving ETH and the 'Mad Max' analogy for OpenAI's dispersal of raw data. Kingy AI offered a more cautious, evidence-based analysis, noting that while public manuscripts and Lean formalizations exist, independent verification has not been completed. Kingy AI also pointed out discrepancies between the quantitative bounds in the natural-language manuscripts and the Lean formalizations, emphasizing that 'there is a Lean proof' does not guarantee every corollary is verified. Both secondary sources agree that the claims are significant but require rigorous scrutiny by the scientific community.

Why it matters

If verified, these results would fundamentally alter fields from computational complexity to particle physics. The ETH refutation could reshape how we understand the limits of algorithmic efficiency, while the Kakeya proof would resolve a long-standing problem in harmonic analysis. The broader significance lies in the new model for AI-generated science: OpenAI is positioning itself as a provider of 'raw research data' rather than a traditional academic author, challenging norms of credit, verification, and the role of AI in scientific discovery.

What to watch

The scientific community will need to independently verify the Lean formalizations and natural-language manuscripts. The physics group is expected to release its own writeup of the quantum field theory results. OpenAI's 'shepherd' model, where external experts guide the conversion of AI outputs into human knowledge, will be tested as a new paradigm for AI-assisted research.