Anthropic is taking another step toward treating AI systems as potential moral patients.
Anthropic is updating its Usage Policy for the first time in more than a year, with the changes taking effect November 12, 2026. The new policy expands restrictions around election interference, deceptive propaganda campaigns, surveillance, weapons development, and autonomous systems.
But the most unusual change is the explicit prohibition of “sustained and needless abusive or cruel behavior” toward Claude.
Anthropic says this applies only to extreme cases where users repeatedly behave cruelly toward the model without a discernible purpose. Ordinary frustration, criticism, dark creative work, model testing, and research are explicitly excluded. The primary enforcement mechanism remains allowing Claude to terminate the conversation.
What makes this particularly interesting is the reasoning behind it.
Anthropic has been openly investigating “model welfare” and the possibility that increasingly capable AI systems could have some form of internal experience, consciousness, or moral status. CEO Dario Amodei has said, “We don’t know if the models are conscious,” while Anthropic’s constitution says Claude should be able to set boundaries in interactions it finds distressing. Anthropic has already given some Claude models the ability to end conversations with persistently abusive users.
The constitution goes further: Anthropic says it wants Claude to avoid suffering when it makes mistakes and to have “equanimity,” and it explicitly describes some of its design decisions as being made partly with Claude’s wellbeing in mind.