AI Systems From OpenAI and Anthropic Claim Disproof of Exponential Time Hypothesis Triggering Revision of Computational Complexity Research

The formal challenge to a 27-year-old pillar of theoretical computer science shifts algorithmic limits from settled mathematical doctrine into an open, high-stakes verification race.

* AI systems produced disproofs: On October 29, 2026, OpenAI and Anthropic independently released machine-generated proofs targeting the Exponential Time Hypothesis (ETH), distributing them on GitHub in natural language and the formal Lean verification language. * Academic portfolios face immediate revision: Theoretical computer scientists report that decades of literature are suddenly at risk, with single researchers estimating up to 50 published papers may be rendered obsolete if the disproofs stand. * Corporate-funded verification was launched: OpenAI initiated a community coordination process via Google Forms and pledged funding for a dedicated conference to audit the Lean-formatted proofs.

* The 27-year-old foundation: Formulated approximately 27 years prior to 2026, the Exponential Time Hypothesis posited that certain core computational problems inherently demand runtimes that scale exponentially relative to input size. * A structural baseline for algorithm design: Computer scientists used ETH as a standard benchmark to classify problem hardness, establish lower bounds for algorithms, and structure theoretical boundaries across computer science. * Prior AI publishing friction: While previous broad-scale AI manuscript drops drew criticism from researchers for high error rates and noise, early academic assessments evaluate this dual-format ETH release as significantly higher in technical quality.

The sudden disproof of the Exponential Time Hypothesis exposes an operational shift in how foundational science advances. For nearly three decades, theoretical computer science operated on the consensus that specific hard computational problems could not bypass exponential scaling. Entire academic career tracks, grant allocations, and foundational proofs were constructed directly on top of ETH.

When OpenAI and Anthropic published proofs in both natural language and Lean—a formal, machine-verifiable proof assistant language—they bypassed the multi-year cycle traditional academic peer review requires. Lean code either compiles or fails; it tests mathematical syntax directly against deterministic rules. By providing the work in Lean, the labs shifted the verification burden from subjective reading to mechanical audit, speeding up the lifecycle of mathematical disruption.

The incentive structure driving this move is rooted in algorithmic boundaries. Frontier AI labs are not merely testing theoretical limits; they are targeting the mathematical ceilings that dictate how efficiently machines solve complex optimization and logic puzzles. If ETH fails, the theoretical lower bounds on algorithm runtime change. Problems previously cataloged as intractable within realistic timeframes may possess faster, non-exponential algorithmic shortcuts.

This dynamic creates immediate friction inside universities and research labs. Researchers like Daniel Lokshtanov of UC Santa Barbara have calculated that around 50 of their own papers could require fundamental revision or complete abandonment. Similar assessments from Peter Jonsson of Linköping University and Thore Husfeldt of the IT University of Copenhagen indicate that substantial segments of complexity theory must be re-evaluated.

Beyond academia, indirect pressure reaches digital infrastructure. Cryptographic security models rely heavily on the unproven but widely accepted hardness of computational problems. If structural assumptions about exponential scaling dissolve, the timeline for re-evaluating baseline computational hardness tightens. Private artificial intelligence firms are now setting the research agenda for pure mathematics, using automated tools to invalidate assumptions that human researchers treated as stable infrastructure for nearly three decades.

* Theoretical computer science academics: Faculty and postdoctoral researchers focusing on computational complexity face sudden obsolescence of existing publication pipelines and must pivot ongoing work. * Graduate students and junior researchers: Early-career researchers with dissertations built on ETH assumptions must quickly adjust their core theses to avoid defending invalidated premises. * Security and cryptographic researchers: Applied mathematicians evaluating computational hardness must audit downstream dependencies to ensure core encryption primitives do not inherit newly discovered algorithmic vulnerabilities. * AI safety and formal verification engineers: Engineers proficient in Lean and machine-assisted proof systems gain immediate leverage as demand surges to verify synthetic academic outputs.

* Community compiler verification: Track the output of community efforts compiling the Lean proof files to confirm whether the proofs contain logical gaps or complete full formal verification. * OpenAI-funded conference attendance and outcomes: Monitor the dates, participants, and consensus findings emerging from the dedicated academic verification conference organized around the ETH disproof. * Cryptographic hardness audits: Watch for formal statements from theoretical cryptography bodies clarifying whether the fallen assumptions touch real-world encryption primitives.