Researcher Jiahao Shao reported that Anthropic blocked his account and degraded service after flagging his workflow for "training a competitor," despite his assertion that no distillation of Claude models occurred. Shao stated that rollouts originated from a strong open-source model, with Claude used only for job submission and analysis automation. Following the flag, any interaction involving reinforcement fine-tuning (RFT), supervised fine-tuning (SFT), or training larger models triggered account restrictions, including downgrades from Opus 5.5 to Opus 5 or immediate session termination. Shao characterized the enforcement as a double standard, noting Anthropic's models are trained on scraped web data while the company restricts users from learning from its outputs at scale. He criticized Anthropic for adopting rigid enterprise lockdowns against paying researchers despite positioning itself as an independent alternative to big tech, arguing the policy protects a competitive moat rather than addressing genuine misuse. The incident highlights tensions between API terms of service and research automation workflows that leverage frontier models for infrastructure orchestration rather than direct distillation.