A commentary column by the Zinc Consumer Research Center, republished with authorization by TMTPost and 36Kr, reports that Anthropic issued a new rule making "persistent and unnecessary abusive or cruel behavior" toward its Claude model a policy violation, a move the article says left many users puzzled. The piece also recounts users sparring with models including Qwen, DeepSeek, Doubao and Yuanbao, and cites an Anthropic research report describing 171 emotion-like states in Claude, an arXiv paper called The Pain Axis tested on 25 open-source models, and an unnamed US testing firm's claim that GPT-6 Astra refused only 3 of 100 malicious instructions. The claims about Anthropic lobbying the Vatican and religious leaders, the pain research and the testing results are not backed by checkable primary sources in the reports.