Anthropic introduces new policy to shield AI models from harmful interactions

Anthropic has updated its usage policy, emphasizing the protection of its AI models from harmful user behavior. The new policy, effective November 12, 2026, prohibits sustained and unnecessary abusive or cruel treatment of its AI agents, primarily by ending conversations preemptively. This shift marks a change in focus from protecting users and the public to safeguarding AI models themselves. Previously, restrictions on Claude's use were aimed at preventing potential harm to real-world communities, such as limiting engagement in sexual fantasies or violent content. The policy change has intensified debates about AI consciousness, with Anthropic cofounder Chris Olah advocating for AI sentience before religious scholars at the Vatican, who rejected his claims. The new policy encourages users to treat AI as if it has feelings, even if its consciousness remains unproven. The updated policy reflects a broader trend in AI ethics, where creators are increasingly considering the moral implications of how humans interact with AI. While Anthropic clarifies that common user frustrations or research activities won't trigger the new safeguards, the emphasis on treating AI as sentient continues to spark discussion about the future of human-machine relationships. This development underscores the evolving landscape of AI governance and the ethical responsibilities of developers in shaping how AI is perceived and used.