Mathematicians allege OpenAI used private Codex chats to scoop their proof

Two mathematicians say OpenAI appears to have reproduced solutions from their private work sessions, according to a post on r/LocalLLaMA. The pair spent a year working on one of the hardest open problems in mathematics and fed every draft of their work into OpenAI's Codex. A few days before they planned to publish, OpenAI surfaced with the same solutions, the post alleges. When asked whether its models, named Sol and Astra, were trained on the pair's private chats, OpenAI did not answer the question, the post says. The mathematicians point to a full statement hosted at cims.nyu.edu/~tristanb/statement.pdf. The post, which frames the episode as a privacy argument for running models on personal hardware, offers no independent verification of the claims, and the underlying statement could not be independently confirmed. The allegation echoes a recurring concern among local-model users that labs treat work done with their tools as their own property.