OpenAI and Anthropic each announced in late October 2026 that internal large language models independently discovered subexponential-time algorithms for counting solutions to k-SAT instances, with claimed running times of roughly 2^O(n log log n / √log n). Both released papers and Lean-formalized proofs on GitHub, prompting a working group led by ETH co-originators Russell Impagliazzo and Ramamohan Paturi plus Ryan Williams to organize public scrutiny. Mechanical verification of the Lean code has succeeded, yet no independent human verdict has certified that the statements refute the standard ETH formulation or its variants like SETH. Traders are watching for any flaw reports, a planned community conference, or formal acceptance that could resolve the market.