OpenAI and Anthropic each announced in late October 2026 that internal large language models independently produced subexponential-time algorithms for k-SAT and 3-SAT, with machine-checked Lean proofs that would refute the Exponential Time Hypothesis if correct. The claims, published on GitHub and preprint servers, drew immediate expert attention because they contradict a foundational assumption used in hundreds of conditional hardness results across complexity theory. A working group led by ETH originators Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams is now conducting an open review, with Lean kernel rechecks already passed but full mathematical validation still pending. Competitive dynamics between the two labs and uncertainty over whether the proofs survive scrutiny are the main drivers of trader positioning, with resolution likely hinging on the outcome of that examination or any peer-reviewed confirmation in the coming weeks.