Anthropic bans being cruel to Claude, from Nov 12

Anthropic rewrote its Usage Policy, the first overhaul in over a year, and one new rule stands out. Users may not engage in "sustained and needless abusive or cruel behavior" toward its models. It takes effect November 12.

The rule is narrow. Anthropic says it targets extreme cases of purposeless cruelty. Ordinary frustration, pushback, dark creative themes and model testing or research are explicitly out of scope. Ending the conversation is the primary enforcement tool, building on the ability Claude Opus 4 and 4.1 got in August 2025. The coverage doesn't detail penalties beyond that, so the viral "throttle or suspend" framing isn't confirmed.

The rest of the rewrite is more concrete. Weapons restrictions now cover the software and components that make weapons work, such as guidance and control systems. The policy bars tracking people without consent, building surveillance tools, and using Claude to recommend who gets investigated or arrested. Deceptive influence campaigns, including hidden sponsors and fake-account amplification, now sit in one section. Anthropic also notes it may sign government contracts that tailor use restrictions.

Why it matters: it's one of the first usage policies to protect the model itself, as well as users and third parties.