Report: Anthropic to ban cruel behavior toward Claude

Anthropic has a new version of its usage policy, which according to Anthropic's own website takes effect on November 12, 2026. The usage policy sets out the rules users must follow when using the company's AI model Claude.

According to Interesting Engineering, the policy prohibits persistent and unnecessary abusive or cruel behavior toward the Claude models. This does not appear in Anthropic's own text, which was reviewed for this article, and the claim rests on that one outlet alone.

According to the coverage, the rule is intended for extreme, repeated cases with no apparent purpose. Ordinary frustration is not covered. Users can still challenge Claude's answers, test the model's limits and explore dark themes in creative work.

The conversation is ended

The primary enforcement measure is ending the conversation. According to Interesting Engineering, Anthropic has not said whether repeated violations could lead to account suspension or other sanctions.

Three tiers of rules

According to Anthropic's website, the policy is built around three tiers: universal standards for everyone, requirements for high-risk use and supplementary guidelines for specific applications.

The prohibitions include illegal activities, attacks on critical infrastructure or unauthorized computer systems, development of weapons as well as biological and chemical agents, violent extremism and terrorism, violations of privacy and child safety, non-consensual intimate material, disinformation, fraud and sexually explicit content.

If Claude is used for recommendations that affect individuals in areas such as health, finance, law, employment, education, housing and benefits, the rules require a qualified person to review the output. Those affected must also be told that AI was used. Physical actions must be governed by independent safety controls. Chatbots and agents must also disclose that the user is talking to an AI and not a human.

Exemption for military and government

According to Interesting Engineering, military and government contracts can be exempted if contractual safeguards sufficiently mitigate harm. This, too, is reported by only one outlet.

It is unclear which parts are new in the 2026 version and which date from the update in August 2025. At that time, Anthropic tightened, among other things, the ban on malicious attacks on computers and networks, while preserving support for legitimate cybersecurity. Anthropic calls the policy a living document that evolves as AI risks evolve.