OpenAI says its AI disproved the Exponential Time Hypothesis, a 1999 computer-science conjecture known as ETH. This has nothing to do with Ethereum. Crypto traders briefly confused the two on social media. The mix-up shows how fast AI math claims can ripple into unrelated markets.
What actually happened
OpenAI published a 67-page paper on 17 October 2025, revised on 25 October, claiming its AI found a deterministic algorithm that solves 3-SAT puzzles in subexponential time, disproving ETH. The company says its model found the disproof on 16 October and formalized it in Lean, a verification tool, the next day. Both sit on GitHub, alongside 722 math manuscripts OpenAI released earlier in October, of which only 162 included a computer-checked result. Computer scientists Russell Impagliazzo, Ramamohan Paturi and MIT's Ryan Williams, who lead the independent review, wrote in an open letter that a result of this significance requires careful, independent scrutiny, according to Decrypt. Anthropic followed on 26 October with a separate paper crediting its Claude model with a similar 3-SAT algorithm.
How we got here
ETH was proposed in 1999 by Impagliazzo, Paturi and Francis Zane. It says 3-SAT cannot be solved much faster than brute force, and hundreds of papers in graph theory, computational biology and databases rely on that assumption, per OpenAI. The claim does not resolve P versus NP, the question of whether every easily checked problem is also easy to solve. Anthropic's Claude formalized Fermat's Last Theorem in Lean in September 2025. Mathematician Terence Tao had already warned that AI tools could flatten open problems before researchers finish them, foreshadowing this overlap between two AI labs.
Why this matters for you
For Ethereum holders, nothing changes. This ETH is unrelated to the blockchain, its token or its roadmap. For AI researchers and builders, the episode tests how fast unverified claims travel and how formal tools like Lean can speed independent checking. OpenAI says it signed no non-disclosure agreements, published raw data and put $5,000,000 (USD) into the Simons Foundation to fund review workshops. That model could become a template for how AI labs release contested proofs. For crypto communities, the confusion is a reminder to verify claims before reacting to headlines that share familiar acronyms.
The bigger question
If AI models can disprove decades-old conjectures before peer review finishes, how should the research community redefine what counts as proof? The reviewers who proposed ETH have not certified OpenAI's claim, leaving open whether machine formalization in tools like Lean can ever replace the slower, human process of scrutiny that mathematics has relied on for centuries.
What to watch
Independent scrutiny from Impagliazzo, Paturi and Williams continues, with no fixed deadline announced. OpenAI's $5,000,000 (USD) Simons Foundation grant will fund workshops and a dedicated conference on the result. Anthropic's Claude paper, dated 26 October 2025, awaits its own review. Crypto traders should watch whether further AI math claims cause similar confusion with token tickers.
