Anthropic faces user allegations regarding training restrictions

A user of Anthropic’s Claude AI has alleged that their account was flagged and blocked for “training a competitor.” The user claims that any activity involving reinforcement learning from human feedback (RFT), supervised fine-tuning (SFT), or the training of larger models resulted in account restrictions or downgrades in model performance.

Anthropic’s existing commercial, consumer, and usage terms explicitly prohibit using its services or outputs to train competing models. While the user maintains that their work involved running analysis on open-source models rather than distilling Claude, Anthropic has not issued a public statement regarding this specific incident.