OpenAI Alleged of Using Mathematicians' Private Chats to Replicate Research

Two mathematicians allege that OpenAI used their private Codex chats and draft research inputs to replicate and publish a major mathematical breakthrough shortly before their planned publication. When pressed on whether private user interactions were used to inform or train its models, OpenAI declined to answer. This situation highlights major intellectual property and privacy concerns surrounding proprietary cloud LLMs, where sensitive research fed into AI assistants could be logged or mined. It reinforces the growing push among researchers and engineers to run open-weight models on local hardware to prevent unauthorized data usage. According to a formal statement by an NYU researcher, the pair spent over a year tackling a complex math problem while sharing intermediate steps with OpenAI Codex. OpenAI unveiled identical solutions right before the researchers' official paper was published, leaving questions open about prompt data retention and training practices.