Researcher Jiahao Shao says that using Claude to help submit training jobs and analyze results led to classification as training a competitor, followed by blocks, terminated sessions and reduced access from Opus 5.5 to Opus 5. He says the training rollouts came from an open-source model and that he did not distill Claude. Related coverage repeats the allegation but does not establish the enforcement criteria or provide official confirmation from Anthropic. Researchers should review applicable terms and workflow boundaries, without treating this individual account as proof that all model research is prohibited.