AI Disproof of Exponential Time Hypothesis Shakes Computer Science Foundations

Key points

• OpenAI and Anthropic announced this week that they have disproven the Exponential Time Hypothesis (ETH), a 27-year-old assumption underpinning much of theoretical computer science.
• ETH posits that the effort to solve specific problems increases exponentially with size; disproving it suggests faster solutions are possible, rendering many existing algorithms and complexity theories obsolete.
• Researchers, including Peter Jonsson of Linköping University and Daniel Lokshtanov of UC Santa Barbara, report that their work may be 'dead' or 'pointless,' though some view the shift as an exciting opportunity to explore new questions.
• OpenAI has published proofs in Lean format on GitHub and commissioned an independent group, including ETH proposers from UC San Diego, to verify the findings, though full understanding is still pending.
• The announcement coincides with other AI-driven breakthroughs in mathematics and physics, such as claims regarding the Navier-Stokes equations and quantum field theories, intensifying debates about AI's role in scientific discovery.

Background

This development follows a series of AI-driven mathematical breakthroughs in 2026, including OpenAI's earlier release of over 700 manuscripts on GitHub in October, which claimed to solve 372 open problems but also included retracted results due to errors. The current ETH claim is part of a broader trend where AI models are rapidly solving long-standing problems in math and physics, leading to what some describe as a 'golden age' of mathematics but also causing existential anxiety among researchers. Previous concerns about AI gatekeeping and the lack of transparency in how these proofs are generated have heightened scrutiny on the current claims.

How outlets are covering it

WIRED emphasizes the psychological impact on researchers, describing the situation as 'doom and gloom' for some, while others like Thore Husfeldt of the IT University of Copenhagen express enthusiasm, comparing the new insights to 'floodlights' in a dark room. The Atlantic highlights the broader crisis in mathematics, with Steven Strogatz noting that AI has fundamentally shifted the field, potentially making human mathematicians obsolete. BBC focuses on the integrity of scientific processes, citing concerns about AI-generated 'slop' and the erosion of trust in peer review, while also noting the potential for AI to accelerate discovery if properly regulated. New Scientist provides a balanced view, acknowledging the 'enormous breakthrough' potential but stressing the need for independent verification and understanding the implications for P versus NP problems. There is a consensus that the ETH claim is significant, but disagreement on whether it represents a net positive or negative for the scientific community, with some fearing the loss of academic norms and others seeing it as a new era of discovery.

Why it matters

The potential disproof of ETH could fundamentally alter the field of computer science, invalidating decades of research on algorithmic complexity and potentially impacting cryptography, which relies on assumptions about computational difficulty. It raises critical questions about the role of AI in scientific discovery, the integrity of peer review, and the future of academic careers. If confirmed, it could lead to more efficient algorithms and new solutions to previously intractable problems, but it also risks undermining trust in scientific processes if AI-generated results are not properly verified. The event highlights the need for robust frameworks to ensure that AI-assisted research maintains scientific rigor and transparency.

What to watch

Independent verification of the ETH proof is underway, with a group led by Russell Impagliazzo and Ramamohan Paturi from UC San Diego examining the findings. OpenAI has committed to funding a conference on the topic and is experimenting with approaches where independent academics organize verification efforts. Researchers are expected to continue unpacking the proof and deducing new results, while also reassessing their own work that relies on ETH. The broader scientific community will likely debate the implications for other fields, including cryptography and theoretical physics, and the need for new norms in AI-assisted research.