OpenAI's AI Models Disprove 27-Year Exponential Time Hypothesis, Rendering Decades of Research Obsolete

Reviewed byNidhi Govil

19 Sources

Share

OpenAI released AI-generated proofs disproving the Exponential Time Hypothesis, a foundational assumption in computer science since 1999. The breakthrough forces researchers to revisit decades of work while raising questions about cryptographic systems and the future role of AI in scientific discovery.

OpenAI and Anthropic Challenge Foundational Computer Science Assumption

OpenAI released AI-generated proofs on October 26 that disprove the Exponential Time Hypothesis (ETH), a 27-year-old mathematical assumption underpinning vast areas of computer science research

1

4

. Hours later, Anthropic released its own refutation of ETH, escalating competition between frontier AI labs

1

. The AI breakthrough came as a complete shock to the scientific community, with University of Bergen professor Fedor Fomin stating that "nobody I know expected it"

4

.

Source: TechCrunch

Source: TechCrunch

The Exponential Time Hypothesis, proposed in 1999 by UC San Diego computer scientists Russell Impagliazzo and Ramamohan Paturi, established limits on how quickly computers can solve certain logical puzzles

1

. The hypothesis suggested that computational effort needed to solve specific problems grows exponentially as problem size increases. OpenAI's AI models devised algorithms solving these puzzles in "sub-exponential" time, potentially proving ETH false

1

.

Massive Implications for Existing Research and Cryptographic Systems

The ramifications extend across theoretical computer science, with UC Santa Barbara professor Daniel Lokshtanov estimating the result renders him "galactically wrong" and makes around 50 of his published papers either "pointless" or requiring revision

2

. Linköping University professor Peter Jonsson described much of his work as "dead" just two weeks before presenting it at a prestigious IEEE Symposium on Foundations of Computer Science

2

.

MIT Professor Ryan Williams, who previously criticized OpenAI's October proof dump as a "DDOS attack on mathematicians," agreed to organize the computer science community's response

1

. Williams identified "two or three alien-looking parts" in OpenAI's proof that require collective examination from the entire scientific community

1

.

Beyond academic concerns, the breakthrough raises immediate cybersecurity questions. An improved algorithm could undermine cryptographic systems that rely on computational difficulty assumptions

1

. Williams pushed OpenAI to fund a conference addressing these implications, emphasizing the need to establish new foundations for computer science research

1

.

Connection to P vs NP and Broader Mathematical Problems

Source: The Verge

Source: The Verge

The Exponential Time Hypothesis relates directly to the P vs NP problem, which questions whether problems with easily verifiable solutions can also be solved easily

1

. While disproving ETH doesn't resolve P vs NP, University of Oxford researcher Leslie Goldberg noted it represents "an enormous breakthrough" if confirmed

4

.

OpenAI verified the ETH result using Lean, mathematical proof-checking software that ensures logical consistency

4

. Stefan Szeider at TU Wien in Austria expects the proof to withstand scrutiny, though researchers must verify the formally proven statement matches its intended meaning

4

.

The announcement accompanied two other major breakthroughs: a solution to the Kakeya conjecture across all dimensions and progress on the Yang-Mills problem, one of seven million-dollar Clay Millennium Problems

3

5

. The Yang-Mills work addresses quantum chromodynamics, validating the mathematical foundations of particle physics

3

.

Source: Scientific American

Source: Scientific American

Scientific Community Grapples with AI-Generated Discoveries

OpenAI structured its releases differently across disciplines this time. Mathematicians received work on GitHub, computer scientists formed a working group requesting public proof sharing and conference funding, while physicists established a group determining release protocols

1

.

The rapid pace of AI-generated mathematical breakthroughs has created psychological strain within academic departments. IT University of Copenhagen algorithms research head Thore Husfeldt reported his department considered hiring an occupational psychologist because "some people are really depressed and really sad" as AI models regularly "annihilate the ongoing research projects of many people"

2

.

Despite the disruption, some researchers express enthusiasm. Husfeldt described the finding as "fantastic," comparing it to "somebody just illuminated an entire subfield in floodlights of knowledge"

2

. University of Sydney researcher Clément Canonne noted that "proving something and understanding why it is true aren't the same thing," emphasizing the need to comprehend the underlying reasoning

4

.

Controversy Over Release Methods and Academic Protocols

The releases contradict recommendations from the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), an expert panel OpenAI organized following mathematician criticism

5

. The large language model producing these proofs remains proprietary, with exact prompts, computation time, and chain-of-thought outputs undisclosed

5

.

OpenAI's approach violated an AGMAI clarifying statement published just one day prior by designating specific experts to shepherd results publicly

5

. The Yang-Mills proof remains unpublished while Nobel laureate David Gross examines it, prompting AGMAI to demand immediate release, stating "there is no scientific reason for delaying the release any further"

3

5

.

The timing proved particularly frustrating for Anthropic mathematician Levent Alpöge and Northwestern University's Ethan Sussman, who were publicly pursuing the Yang-Mills problem independently

3

. They updated their GitHub repository with intermediate results the night before OpenAI's announcement, believing they were "maybe even as close as a few weeks away" from completion

3

.

Royal Holloway's Magnus Wahlström described the phenomenon as "mathematics done by press," expressing concern that proprietary AI models can suddenly overturn months of researchers' work

4

. OpenAI commissioned an independent group including ETH co-proposers Impagliazzo and Paturi to examine findings, with their open letter acknowledging the work could "change an important part of theoretical computer science"

4

.

© 2026 TheOutpost.AI All rights reserved