A high-profile announcement of scientific breakthroughs by OpenAI has ignited widespread frustration across the academic community, as researchers push back against what they describe as gatekeeping and disruptive disclosure practices by the artificial intelligence laboratory.
On Monday, OpenAI unveiled purported solutions to three landmark problems spanning mathematics, physics, and computer science. The company claimed its models resolved the geometric Kakeya conjecture across all dimensions, refuted theoretical computer science’s long-standing Exponential Time Hypothesis, and produced a mathematical proof addressing the existence portion of quantum Yang-Mills theory, a cornerstone of particle physics and one of the seven Millennium Prize problems.
Dispute over access and handpicked reviewers
Rather than winning unreserved praise, the announcements renewed tensions that have brewed for weeks between frontier artificial intelligence labs and academic researchers. While OpenAI shared its raw output for the mathematics problem on GitHub, it convened a panel of outside computer scientists to examine the algorithm claims and chose to withhold the physics proof entirely until an external group decides how to publish it.
The decision to control which scholars inspect the underlying data drew swift rebukes from academic organizations and university faculty, The Verge reported. Just one day before the announcement, the Advisory Group on Mathematics and Artificial Intelligence warned that selecting specific individuals to receive outputs creates unequal access and risks compromising academic independence.
Following the announcements, the advisory group urged OpenAI to immediately release the natural-language and Lean formalization outputs for its physics result, noting that the company had already completed computational verification and there remained no scientific basis for withholding the data.
“The fact that they handpick the group is a disaster,” Andreas Karch, a physics professor at the University of Texas at Austin, told The Verge. “The large labs have enough power as it is. They should not be picking winners and losers.”
Competition between frontier labs
The disclosures also highlighted bitter rivalries between technology companies competing for supremacy. Scientific American reported that OpenAI’s Yang-Mills announcement effectively scooped independent, publicly documented work by Northwestern University mathematician Ethan Sussman and Levent Alpöge, a mathematician employed by OpenAI rival Anthropic. The duo had updated a public repository with their own progress just hours prior to OpenAI’s statement.
Hours after OpenAI claimed to refute the Exponential Time Hypothesis, Anthropic announced that its own models had also refuted the theory, TechCrunch reported. The hypothesis, formulated in 1999, holds that certain complex logic problems cannot be solved substantially faster than through brute-force computation.
Massachusetts Institute of Technology computer science professor Ryan Williams, who was asked by OpenAI to help interpret the algorithmic results alongside the researchers who originally framed the hypothesis, emphasized that taking part did not constitute an endorsement. Williams noted that while portions of the computer science proof appear routine, other parts appear “alien-looking” and demand broader scrutiny, according to TechCrunch. He also noted that preliminary checks suggested the potential algorithm would not compromise existing cryptographic systems.
The struggle to establish new academic norms
The latest dispute follows an earlier episode at the start of October, when OpenAI released hundreds of machine-generated preprints that flooded mathematical research channels, an incident Williams previously likened to a distributed denial-of-service attack on mathematicians.
OpenAI has said its shifting distribution strategies represent an effort to experiment with how best to share findings, with company leaders consulting outside experts to navigate the trade-offs. However, many scientists argue that privately selecting collaborators does not substitute for transparent peer evaluation.
“It doesn’t make any sense that first only a few distinguished scientists can look at them,” Constantin Kogler, a researcher at the Institute for Advanced Study, told The Verge.
Whether OpenAI will yield to calls from academic bodies to release its withheld proofs to public scholarly repositories remains unconfirmed, even as mathematicians and computer scientists begin the grueling task of verifying the work that has been made public.
