Another mathematician questions whether OpenAI 'borrowed' unpublished research! German mathematician Andreas Thom has publicly questioned whether OpenAI's flagship model, Astra, used his unpublished research to prove a key result. He even shared email exchanges between the two parties. This incident is separate from the previously reported Buckmaster-Alpöge fluid dynamics proof dispute, but the circumstances are strikingly similar. Andreas Thom pointed out that OpenAI's Astra model solved a 27-year-old unsolved problem in group theory, with a critical step that appears to be derived from his unpublished research with collaborators—research he had been working on for months using ChatGPT to explore the same problem. He wrote to OpenAI researchers asking whether his conversations had been used for training or problem-solving, but the response he received was a vague denial: "It didn’t happen," with no further explanation provided. Thom noted that the reply conflated two distinct questions he had raised. Moreover, weeks later, in another similar dispute (the Buckmaster-Alpöge case), OpenAI's response included a disclaimer that "de-identified data cannot be ruled out as being used to improve the model," yet this disclaimer was conspicuously absent in his case. He emphasized that if non-public user research is being used to improve models and then leveraged to publish results ahead of the original researchers, it raises serious ethical concerns.