Mathematicians Challenge OpenAI Over Math Training Data and Chat Logs

Concerns over AI training data are shifting from published web scrapes to private, unpublished research and conversational prompts. Mathematician Andreas Thom and other researchers have publicly challenged OpenAI, demanding proof that recent mathematical milestones were not powered by uncredited interactions with ChatGPT. The dispute follows growing tension over whether frontier models are quietly absorbing novel academic workflows. As AI labs withhold their data pipelines, researchers find that the boundary between testing a model’s reasoning and inadvertently donating intellectual labor has grown increasingly ambiguous.