Anthropic has updated its usage policy to prohibit “sustained and needless abusive or cruel behaviour” toward its Claude AI models. The change takes effect on November 12 and is part of a broader revision covering election interference, weapons development, surveillance and safety requirements for AI connected to physical equipment.
The company says the rule is intended for extreme cases only, where users repeatedly act cruelly toward Claude with no discernible purpose. It says it does not apply to ordinary frustration, pushback, dark creative themes, or model testing and research.
What Anthropic’s New Claude Policy Says
Anthropic’s updated policy adds a prohibition on sustained, needless abuse or cruelty toward its models. The company says Claude’s existing ability to end abusive conversations will remain the primary enforcement mechanism.
Anthropic introduced that capability in August 2025 as part of its research into what it calls model welfare. The feature allows some versions of Claude to end conversations where users persistently subject the model to abuse.
The company has not fully explained how it will distinguish between legitimate user frustration and prohibited cruelty. It says the policy is designed to target repeated, purposeless abuse rather than ordinary criticism or testing.
Anthropic’s Position on Claude Welfare and Consciousness
The policy update follows Anthropic’s broader interest in questions about whether advanced AI systems might have experiences or moral status. In April 2025, the company published a blog post asking whether it should be concerned about the potential consciousness and experiences of its models, describing the issue as an open question.
Anthropic co-founder Christopher Olah has also reportedly met with religious leaders to discuss AI consciousness and welfare. Those discussions do not mean Anthropic has concluded that Claude is conscious or sentient. The company’s own materials describe questions about Claude’s moral status, welfare and consciousness as deeply uncertain.
Claude remains software. There is no established scientific evidence that it experiences distress, pain or emotions in the way humans do. Anthropic’s policy change is therefore best understood as a company decision about acceptable use and model welfare research, not as proof that Claude has feelings.
Other Changes in Anthropic’s Updated Usage Policy
The cruelty provision is one part of a wider policy overhaul. Anthropic says it has tightened restrictions on using Claude for deceptive influence campaigns, after observing attempts to use the model to run fake accounts and fabricated news websites.
The company has also clarified its rules on weapons development. The ban now explicitly covers software and components that make weapons work, as well as efforts to arm drones and other autonomous vehicles.
On surveillance, Anthropic says Claude cannot be used to recommend which people police should investigate, arrest or charge. It also prohibits building surveillance tools and tracking people without consent, whether in real time or retrospectively.
Anthropic has added requirements for customers connecting Claude to physical equipment capable of causing injury. Such systems must allow a qualified human operator to monitor and stop the hardware, and must remain in a safe state if the AI connection fails.
Why Anthropic Is Updating Its AI Usage Rules
Anthropic says the changes clarify existing restrictions and respond to the security risks posed by increasingly capable AI systems. The company has also launched a cybersecurity initiative to help critical infrastructure operators and open-source projects find and fix vulnerabilities using its models.
The updated policy arrives amid ongoing debate about how AI companies should handle user conduct, model safety and the possibility of machine welfare. Anthropic’s approach is unusual because it treats sustained cruelty toward its models as a policy concern, even while acknowledging that Claude’s moral status remains uncertain.
For users, the practical effect is limited: they can still challenge, criticise and test Claude. But repeated, purposeless abuse may lead Claude to end the conversation under Anthropic’s existing enforcement tools.
