OpenAI says its AI disproved the Exponential Time Hypothesis

Originally published on ColdCaseTech.

Disclosure: this post contains affiliate links. We may earn a commission at no extra cost to you.

Then vs. Now (News report): Then: OpenAI announced on 2026-10-26 three new discoveries in mathematics, computer science and physics, including a claimed proof that the Exponential Time Hypothesis is false. Now: OpenAI published the math and computer science materials on GitHub, while the physics group has decided to hold off on releasing the AI-produced artifacts.

On 2026-10-26, OpenAI announced three new discoveries in mathematics, computer science and physics, concerning harmonic analysis, complexity theory and quantum field theory (OpenAI). The company states that its models proved the local smoothing conjecture in every dimension and that its system proved the Exponential Time Hypothesis (ETH) false (OpenAI). It reached out to external experts and asked them to act as shepherds, organizing groups to understand the new results (OpenAI).

Key Facts

Timeline

• 2026-10-06: OpenAI made an earlier release of math output, the strategy it followed again for the local smoothing result (OpenAI).
• 2026-10-18: OpenAI obtained the AI output for the math result, which was formalized in Lean the following day (OpenAI).
• 2026-10-26: OpenAI announced the three discoveries and added the results to GitHub (The Verge).

Details (as Reported)

OpenAI describes the manuscripts as a form of raw research data and states that the shepherds advised three different release strategies (OpenAI). For the math result, the AI output was released and researchers are free to organize their own efforts to explain it (OpenAI). On the computer science result, an external group explains the work while the AI output stays public (OpenAI). In physics, an external group defers to its own judgment on releasing materials (OpenAI).

The Verge reported that the ETH paper gives a deterministic algorithm that exactly counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space, and that the 3-SAT case refutes the hypothesis (The Verge). OpenAI notes that ETH has been used in hundreds of papers to derive conditional hardness results, and that finding better algorithms will require significant further research (OpenAI). Impagliazzo and Paturi are original proposers of the hypothesis, and Williams has refuted strong versions of it (OpenAI). The group asked that all raw data be made public, and OpenAI agreed to fund a conference on the implications (OpenAI).

The collaborations follow four principles: transparency, no academic credit for OpenAI, recognition for the groups, and independence (OpenAI). Participants sign no nondisclosure agreements, and participation is voluntary and unpaid (OpenAI).

Engadget reported that the announcement follows a release of manuscripts with over 700 pages covering 372 major math problems, and that the physics group provided an email address for questions about its efforts (Engadget).

Turn text into natural-sounding speech. ElevenLabs generates voiceovers and narration from text, in many voices and languages. Try ElevenLabs (affiliate link)

What Is Not Yet Confirmed

• OpenAI states that it does not yet know which of the three release approaches will be most useful (OpenAI).
• The ETH work is still to be verified and explained by the Impagliazzo, Paturi and Williams group (OpenAI).
• The physics information has not been released in full (The Verge).
• The AGMAI panel of math advisers recommended that AI companies not adopt the practice of designating certain people to receive output first, citing unequal access, borrowed reputation and biased selection (The Verge).

Sources

• OpenAI, 2026-10-26
• The Verge, 2026-10-26
• Engadget, 2026-10-27

Tools related to this case

• A cheap, disposable Linux box for testing and research: Contabo VPS
• Masked email addresses so one leak does not expose your real inbox: Surfshark Alternative ID
• Phishing-resistant sign-in for email, cloud and exchange accounts: YubiKey hardware security keys

These are affiliate links: we may earn a commission at no extra cost to you. As an Amazon Associate, ColdCaseTech earns from qualifying purchases.

ColdCaseTech is an independent record of technology's promises: every line links to its source. New cases daily at coldcasetech.com. Follow on X, Bluesky or Telegram.