Claude independently discovered an algorithm that refutes the exponential time hypothesis. Anthropic’s internal Claude research model, without human guidance, independently discovered a new algorithm capable of solving the 3-SAT problem faster than previously thought theoretically possible. 3-SAT is a classic problem that determines whether a set of logical constraints can be simultaneously satisfied. For decades, researchers believed such problems could not be significantly accelerated—this belief is known as the exponential time hypothesis. If this new algorithm holds, it would mean the hypothesis is invalid. This development demonstrates that artificial intelligence can now autonomously propose significant algorithms and provide verifiable proofs for core problems in theoretical computer science. If the results withstand scrutiny, they could reshape long-standing perceptions about the difficulty of certain computational problems. The paper and proof have been made publicly available for further validation. The entire process shows that the model can independently achieve high-level theoretical discoveries, though final refinement still requires human oversight.