The new policy, effective November 12, targets repeated abuse with no clear purpose and also clarifies rules on weapons, surveillance, elections, and robotics.
What You Need to Know
- Anthropic bans cruelty toward Claude but does not consider ordinary frustration to be abuse — complaining about bad responses, persisting with a request, writing dark fiction, or testing the model are not covered by the ban.
- The rule takes effect on November 12, 2026, and targets repeated abusive behavior without any discernible purpose. The primary enforcement mechanism is allowing Claude to end conversations with persistently abusive users.
- The question behind the policy is unsettling: Anthropic admits it does not know whether its models are conscious, yet it is already banning certain forms of cruelty toward them.
What Anthropic’s New Policy Bans
Anthropic has added a prohibition on sustained, gratuitous abusive or cruel behavior directed at its artificial intelligence models, such as Claude, to its 2026 Usage Policy.
The rule, published on October 8, takes effect on November 12, 2026, and is aimed at extreme cases in which users repeatedly act cruelly toward the models with no discernible purpose.
What Does NOT Count as Abuse
According to the company, the restriction does not apply to:
- Everyday frustration with poor responses;
- Pushing back or repeatedly pressing a request;
- Dark creative themes, such as fiction and role-play;
- Testing and research involving the models.
In other words, venting frustration over a bad answer or pressing a request does not, by itself, violate the rule.
What Happens to Violators
In August 2025, Anthropic announced that Claude could end conversations with persistently abusive users on Claude.ai and in Claude Code. The company says this will remain its primary enforcement mechanism for the rule.
Anthropic has not clarified whether it will take additional, specific measures for this type of behavior, such as user bans. Its general policy, however, provides for actions such as warnings, restrictions, suspension, or termination of access in the event of violations.
The Other Policy Changes
The update also clarifies restrictions intended to reduce the risks associated with AI use:
- Deceptive campaigns: Bans concealing who is behind messaging or artificially amplifying its reach with fake accounts and fabricated websites, in both political and commercial activities.
- Elections: Prohibits misleading voters, spreading false information about candidates or voting, and attempting to suppress participation. Legitimate civic uses, such as translating election information, remain allowed.
- Weapons: Clarifies that the ban covers software and components essential to a weapon’s operation, as well as actions such as arming drones and other autonomous vehicles. The company reported attempts to use Claude to develop weapon guidance and control software.
- Surveillance: Bans tracking people without consent and using Claude to recommend who should be investigated, arrested, or charged. Anthropic says the revision makes the restrictions more explicit without necessarily changing how they are enforced in practice.
- Health and finance: Clarifies requirements for high-risk uses that may affect rights, financial resources, or access to essential services. In those cases, a qualified human must provide oversight, and the affected person must be informed that AI is being used.
- Robotics: Requires that a qualified operator be able to monitor and shut down autonomous equipment capable of causing injury. The equipment must also remain in a safe state if Claude is disconnected.
There is one caveat: Anthropic may adapt certain restrictions in contracts with specific government customers, provided it deems the safeguards sufficient to mitigate the risks.
Why Is the Company Doing This?
The ban on abusive behavior is tied to Anthropic’s research on the welfare of its AI models. The company believes that possible internal experiences and questions about consciousness deserve investigation as these systems evolve. CEO Dario Amodei acknowledged the uncertainty: “We don’t know whether the models are conscious.”
Microsoft AI, for its part, takes a different stance in its code of conduct. Microsoft AI CEO Mustafa Suleyman rejects the idea that AIs should have legal personhood or deserve rights and welfare of their own. The revised version of the code of conduct, however, acknowledges that the science on artificial consciousness is still inconclusive.
What This Means for You
For the average user, the change does not turn every flash of irritation into a violation. The rule targets repeated abusive or cruel behavior with no discernible purpose — not everyday frustration, pushback, dark creative themes, or testing and research.
The most striking aspect is that one of the world’s largest AI companies now formally prohibits certain forms of cruelty toward its own models, while the question of whether AI can be conscious remains unresolved.
Sources and References
- Anthropic — 2026 Usage Policy
- The Verge — Anthropic’s new usage policy addresses abusive behavior toward Claude
- Anthropic — Research on ending a subset of conversations