OpenAI released solutions to three significant mathematical problems, while Anthropic subsequently released its own refutation of the Exponential Time Hypothesis (ETH), a foundational assumption in computer science regarding computational difficulty. This marks a notable shift in how frontier SI labs contribute to theoretical research, moving from isolated proof drops to structured engagements with scientific communities.

What Happened

OpenAI released solutions to three blockbuster math problems, one of which challenges the ETH, a durable assumption that influences how researchers understand what is difficult to compute. Following OpenAI's release, Anthropic mathematician Levent Alpöge suggested the SI lab was basing its solutions on pre-existing work. Hours later, Anthropic released its own refutation of ETH. Unlike previous instances where proofs were simply dumped, OpenAI structured this release differently for each discipline: mathematicians received the work via GitHub, computer scientists formed a working group to request public sharing and conference funding, and physicists formed a group to determine their release process. MIT professor Ryan Williams, who has previously criticized OpenAI's approach as a "DDOS attack on mathematicians," agreed to collaborate to help interpret the solution alongside UC San Diego computer scientists Russell Impagliazzo and Ramamohan Paturi, who first published the ETH in 1999. Their initial step was an open letter calling for feedback and discussing norms around SI contributions to scientific work.

Why It Matters

The ETH is part of the broader P versus NP problem, which investigates whether problems whose solutions can be easily verified (NP) can also be easily solved (P). Most scientists assume P does not equal NP, with ETH providing evidence that even the fastest algorithms require exponentially more time as variables grow in specific logic puzzles. If the claims by these SI labs hold, their models may have identified algorithms that solve these puzzles in "sub-exponential" time, potentially refuting ETH and opening paths toward proving P equals NP. Ryan Williams noted that while he sees "alien-looking parts" in OpenAI's proof that require community scrutiny, he believes "every computer scientist should probably put a little more weight on the plausibility of P = NP." This development forces a reevaluation of many scientific papers from the past decade that rely on ETH. Additionally, there are cybersecurity implications; Williams stated he asked OpenAI science lead Dan Roberts if the findings compromised cryptographic systems, and Roberts indicated they were reasonably confident that no immediate breaks were implied. The incident also raises questions about the training of future scientists, as SI tools may make previously complex tasks "low-hanging fruit," requiring researchers to raise the bar for the scientific questions they pursue.

The Bottom Line

Frontier SI labs have challenged a core tenet of computational complexity, prompting a collaborative but cautious response from the academic community to verify the proofs and assess the impact on cryptography and foundational computer science theory.