Mathematicians Accuse OpenAI of Using Private Research to Scoop Discoveries
Researchers allege the AI giant used private chat logs and unpublished ideas to solve longstanding mathematical problems.
Two mathematicians have accused OpenAI of using unpublished research and private AI interactions to achieve major mathematical breakthroughs, sparking a debate over intellectual property in the age of generative AI. The allegations suggest the company may have leveraged private user data to beat academic researchers to career-defining proofs.
According to reports from TechCrunch, The Verge, and Notebookcheck, the controversy involves two distinct cases. Mathematician Andreas Thom alleges OpenAI was "plainly dishonest" regarding whether his private ChatGPT sessions informed a proof concerning non-sofic groups. Separately, Tristan Buckmaster claims OpenAI used its massive compute resources to "race" him to a solution for the Navier-Stokes Millennium Prize problem after hearing rumors of his progress, potentially benefiting from his private Codex logs.
OpenAI recently announced solutions to 10 longstanding mathematical problems using its "Astra" model, including the construction of a non-sofic group. For the Navier-Stokes problem, the company claimed it utilized 10,000 AI agents and approximately $22.5 million in compute, generating roughly 300 billion output tokens to reach the solution. OpenAI has denied the allegations, stating its researchers and agents did not see the mathematicians' work until it was released publicly and that no specific user data was accessed to solve the problems.
The Battle for Priority
This dispute highlights the tension in the "black box" nature of AI training data. In mathematics, the priority of discovery—being the first to publish a proof—is the primary currency of professional success. Many researchers use tools like ChatGPT and Codex to brainstorm or refine ideas; however, if these interactions are absorbed into training sets, the AI may effectively "regurgitate" unpublished intellectual property.
Thom argues that standard data protections are insufficient for this work, stating that "de-identification may remove a name; it does not remove the intellectual content of a mathematical idea." He described OpenAI's denials as "dishonesty to say the least."
Industry Implications
If AI companies use private researcher data to scoop academic publications, it could create a significant "chilling effect" across academia. The power imbalance is stark: while individual mathematicians rely on peer-reviewed publication for their careers, tech giants possess the compute power to iterate through millions of possibilities in a fraction of the time. This may discourage mathematicians from using AI tools or sharing early-stage breakthroughs for fear of being scooped by a well-resourced corporation.
What's Next
As the debate intensifies, the focus shifts to whether AI companies can provide transparent proof of their training data provenance. The mathematical community is now calling for greater accountability to ensure that the tools designed to assist discovery do not become instruments for intellectual theft. Whether these allegations lead to formal legal challenges or a shift in how AI companies handle private researcher data remains to be seen.