OpenAI and Anthropic each announced on October 26, 2026, that internal large language models independently discovered subexponential algorithms for 3-SAT and #3-SAT, which would refute the Exponential Time Hypothesis if verified. The claims include Lean-formalized proofs posted on GitHub, with OpenAI’s dated to an October 16 model output. Original ETH authors Russell Impagliazzo and Ramamohan Paturi, along with Ryan Williams, are leading an open community review process that includes public discussion and a planned conference. The results remain unpeer-reviewed, and experts note the theoretical nature of the bounds while highlighting their potential to upend hundreds of conditional hardness results in fine-grained complexity. Traders are watching the verification timeline for any resolution signals.