AI may have just challenged one of computer science’s most fundamental assumptions. But the bigger story isn’t ETH. On October 26, OpenAI announced that an internal AI model had produced a claimed refutation of the Exponential Time Hypothesis (ETH) — a foundational conjecture underpinning hundreds of computational hardness results. The claim: A deterministic, subexponential-time algorithm for exact counting of satisfying assignments in fixed-width Boolean formulas. If correct, ETH falls. So do several stronger conjectures. But P ≠ NP remains unresolved. And theoretically faster doesn’t necessarily mean practically useful. The proof has been formalized in Lean. An independent examination is being organized by Russell Impagliazzo, Ramamohan Paturi, and MIT’s Ryan Williams. Their involvement is not an endorsement. Independent validation remains essential. My perspective: We’re potentially witnessing three breakthroughs, not one. Mathematical: A longstanding computational complexity barrier may have been overturned. Scientific: AI may be transitioning from assisting researchers to generating genuinely novel knowledge. Institutional: The scientific community must rethink how discoveries are verified, understood, credited, and reproduced. The third may prove the most consequential. For centuries, generating frontier-level discoveries required scarce human expertise and years of research. What happens when AI can generate credible mathematical discoveries faster than humans can independently validate and understand them? The bottleneck shifts from discovery to verification, interpretation, and scientific judgment. And eventually, AI may accelerate those processes too. The real inflection point isn’t AI solving one famous problem. It’s AI repeatedly expanding the frontier of human knowledge. We aren’t there conclusively yet. But this development deserves serious attention. The question is no longer simply how intelligent AI can become. It’s whether our scientific institutions can evolve fast enough to evaluate what AI discovers. OpenAI announcement: https://t.co/G0p0VEMjlC Research papers and Lean formalizations: https://t.co/jVif4w5JKB #ArtificialIntelligence #ScientificDiscovery #OpenAI #ComplexityTheory #AIResearch #ComputerScience #FutureOfAI #AgenticAI
