An OpenAI artificial intelligence model has refuted the Exponential Time Hypothesis (ETH), overturning a 27-year-old theoretical computer science assumption first proposed by Russell Impagliazzo and Ramamohan Paturi in 1999. However, experts confirm that the algorithmic discovery poses no threat to the cryptography securing Bitcoin (BTC) and Ethereum.
- OpenAI generated a natural-language proof on October 16, formalizing it in Lean software on October 17 to refute the Exponential Time Hypothesis (ETH).
- The new algorithm solves 3-SAT logic puzzles in roughly 2^(n/√log n) steps for n variables, beating the exponential floor predicted by ETH.
- OpenAI President Greg Brockman and UT Austin professor Scott Aaronson noted the discovery does not prove P = NP, leaving Bitcoin encryption safe.
- UC San Diego researchers Impagliazzo and Paturi, alongside MIT professor Ryan Williams, are leading the review of the machine-verified proof.
Breaking the Exponential Floor Without Threatening Bitcoin
OpenAI's AI system produced a natural-language proof disproving ETH on October 16, which researchers formalized the following day using Lean proof-verification software. The newly discovered algorithm processes 3-SAT logic puzzle inputs of n variables in approximately 2^(n/√log n) steps. While this runtime breaks the strict theoretical floor established by Impagliazzo and Paturi in 1999, OpenAI cautioned that the algorithm remains far too slow for practical real-world computational use.
The refutation generated widespread interest across both computer science and crypto communities concerned about long-term quantum risks to Bitcoin's cryptography. University of Texas at Austin computer science professor Scott Aaronson recounted hearing the news on his Shtetl-Optimized blog after students alerted him to the finding. Formal peer review of the proof is currently underway, led by original hypothesis authors Impagliazzo and Paturi of UC San Diego alongside MIT's Ryan Williams.
P Versus NP and the Security of Digital Assets
Crucially, disproving ETH does not resolve the broader P versus NP problem. On October 26, 2026, OpenAI President and Co-founder Greg Brockman clarified the situation on social media, posting that "p still probably doesn't equal np" while announcing three major scientific discoveries across harmonic analysis, complexity theory, and quantum field theory.
Because ETH is a stricter condition than P ≠ NP, disproving it does not break elliptic curve cryptography. Benjamin Cowen, a member of BeInCrypto’s Market Intelligence Expert Council, emphasized that the refutation does not mean Bitcoin or Ethereum networks are compromised. Beliefs about mathematical impossibility have fallen before—such as Anatoly Karatsuba disproving Andrey Kolmogorov's multiplication conjecture in 1960 or Volker Strassen finding a faster matrix multiplication method in 1969. Cryptographer Matthew Green has similarly warned that rapid AI progress could eventually test cryptographic limits.
Why It Matters
This breakthrough highlights how rapidly AI systems are beginning to challenge foundational assumptions in theoretical computer science. While Bitcoin's cryptographic architecture remains completely safe today, the ability of AI models to discover unexpected math shortcuts compresses the timeline for evaluating cryptographic resilience. As artificial intelligence advances, digital asset security will increasingly depend on proactive upgrades to encryption standards long before practical vulnerabilities emerge.
