Anthropic's new usage policy bans "sustained and needless" abuse of Claude, from November 12

Anthropic has published a rewritten Usage Policy that takes effect on November 12, 2026, its first since the version that took effect on September 15, 2025. One new line, among the rules everyone must follow, prohibits users from engaging "in sustained and needless abusive or cruel behavior toward our models". We compared the two versions on Anthropic's site line by line.

WHAT IS NEW

- Abuse of the models: the line above, which the 2025 policy did not have. - Weapons: new bans on software for targeting, fire control or engagement, and on putting weapons onto drones, vehicles or other unmanned or autonomous platforms. The existing rules on chemical, biological, radiological and nuclear weapons are reworded and broadened. - Influence campaigns: a new ban on hiding who is behind content meant to sway public opinion, officials or other decision-makers. - Surveillance: a new ban on identifying, unmasking or locating people from anonymous or pseudonymous activity, or compiling dossiers on them, in order to harass, intimidate, surveil or harm them. - Scope: the extra guidelines now cover consumer-facing chatbots and agents, and software listed in Anthropic's directories, where the 2025 text named Model Context Protocol servers.

HOW IT IS ENFORCED

Anthropic says its Safeguards Team may warn a user, or throttle, limit, suspend or end their access. The policy also says that being blocked by one of its real-time safeguards does not by itself mean a user broke the rules.

WHAT WE DO NOT KNOW

How Anthropic will decide that abuse toward a model is "sustained and needless". The Verge, which reported the update first, says ending the conversation remains the main tool; the policy text itself does not say.