OpenAI and Anthropic announced breakthroughs this week claiming to have disproved the 27-year-old exponential time hypothesis (ETH), an foundational assumption that has underpinned vast swaths of theoretical computer science, complexity theory, and algorithm design. According to Wired, both companies published their machine-checkable proofs on GitHub in natural language and the Lean format, triggering immediate disruption across academic institutions globally just weeks ahead of a major IEEE conference.
The exponential time hypothesis has long served as the baseline assumption that solving certain computational problems requires an amount of effort that scales exponentially with problem size. With AI models successfully challenging this premise, researchers warn that numerous published papers and ongoing projects may now be rendered obsolete or in need of immediate revision. Academic response has ranged from shock and concern over the psychological toll on researchers to cautious enthusiasm for new avenues of exploration.
Immediate Impact on Academic Research and Upcoming Conferences
Linköping University professor Peter Jonsson, whose work is scheduled to be presented in two weeks at the IEEE Symposium on Foundations of Computer Science, noted that much of his recent research is effectively “dead” if the AI-generated proofs hold up. Jonsson expects this dilemma to affect numerous attendees at the conference, predicting researchers will open presentations by acknowledging that ETH is false and portions of their work no longer function. Similarly, Thore Husfeldt, head of the algorithms research group at the IT University of Copenhagen, described the development as a once-in-a-generation shift that instantly invalidated much of his recent subfield work.
Despite the disruption, some mathematicians and computer scientists have welcomed the clarity, comparing the sudden illumination of a complex subfield to turning on floodlights in a dark room. However, the sheer volume of AI-generated manuscripts—such as a 17-page paper received by Husfeldt from a random Gmail account, alongside OpenAI’s recent release of 722 manuscripts—has overwhelmed researchers trying to manually parse and verify complex mathematical proofs.
Verification Efforts and Future Security Implications
Because researchers have not yet independently verified or fully digested the complex AI-produced proofs, OpenAI has enlisted the original authors of the hypothesis from UC Santa Barbara, alongside an MIT academic, to rally volunteers to review the GitHub documentation and provide feedback. UC Santa Barbara professor Daniel Lokshtanov estimated that the result could make around 50 of his published papers obsolete, calling the development a “giant bomb dropped into the middle of the field.”
Beyond theoretical computer science, experts note that while the current ETH disproof does not immediately break existing cryptography—which relies on the assumption that specific problems take impractically long to solve—cryptosystem designers should take notice. As academic institutions organize review efforts and plan dedicated conferences funded by OpenAI, the broader scientific community faces the challenge of adapting its methodologies to a landscape where artificial intelligence can rapidly reshape foundational mathematical assumptions.