Who gets to check OpenAI’s science?
Researchers want access to the work behind OpenAI’s latest scientific claims. Expert involvement is a starting point for scrutiny, not a certificate of correctness.
OpenAI’s latest scientific claims raise a question that matters beyond mathematics: who gets to examine the evidence? The company announced results across mathematics, computer science and physics on October 26. Researchers are now debating both the findings and the terms on which the wider community can investigate them.
The company’s release describes its model-generated manuscripts as raw research material. It is trying three approaches: publish the mathematics output; publish the computer-science output alongside an external working group; and leave release of the physics materials to a selected group of scientists. OpenAI says participants can discuss their work and publish conclusions without nondisclosure agreements.
These distinctions matter. An announced result, a downloadable argument and an independently understood result are different stages of research.
Expert involvement is not certification
The computer-science claim concerns the Exponential Time Hypothesis, an assumption about how quickly certain logical problems can be solved. Its examination is being organized by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams.
Their message to the research community explicitly warns that their involvement does not certify the claim. They ask researchers to inspect the argument, identify limitations, share corrections and distinguish established consequences from possibilities. They also ask how checking and explaining results should receive credit when AI makes producing them faster.
That is a useful way to read the announcement: there is substantial work available to examine, and an open effort to understand it. We have not independently verified the proofs.
The access dispute
The Advisory Group on Mathematics and Artificial Intelligence has called for immediate publication of the physics outputs. Its objection is to unequal access and company influence over who gets an early opportunity to work on a result. The group respects the scientists involved while arguing that understanding should develop openly.
The Verge’s October 27 reporting finds excitement about the scientific possibilities alongside frustration with restricted access. OpenAI itself acknowledges that selecting initial collaborators creates information asymmetries. As of this review, its release page still says the physics group is holding back the underlying artifacts while preparing its own writeup.
For readers, three questions help keep the story grounded: Can outsiders obtain the underlying work? What have independent researchers actually checked? Can they publish criticism and corrections freely?
Our view: making research understandable, contestable and widely accessible is part of realizing its value. Enthusiasm for a potentially important discovery should increase the demand for that work. Our Harms & evidence reading guide offers more ways to distinguish a claim from the evidence supporting it.