Researcher Claims Anthropic Unjustly Banned Account Over Model Training Dispute

Researcher Jiahao Shao stated on X that Anthropic wrongly banned their account and restricted access under the accusation of training a competitor model. Shao clarified that they never distilled Claude, but instead used it only to help submit jobs and run analysis for rollouts from an open-source model. This incident highlights growing industry concerns regarding the transparency, strictness, and potential overreach of automated account banning practices and usage policies implemented by major AI labs. It underscores the friction between closed-source AI providers and independent researchers working on open-source training methodologies. Shao noted that following the ban, any activity touching on Reinforcement Fine-Tuning (RFT), Supervised Fine-Tuning (SFT), or training larger models results in automated blocking. The dispute centers on whether utilizing Claude as an auxiliary assistant for open-source workflows violates service terms regarding competitive model development.