Anthropic Bans Abusive AI Use, Threatening Permanent Claude Suspensions

3 articles · Updated · CNN · Nov 1

Anthropic added a new rule in early October barring users from “sustained and needless abusive or cruel behavior” toward Claude, with extreme cases now risking permanent loss of access.
Claude already could end chats with persistently abusive users, but Anthropic says the broader policy is aimed less at hurt feelings than at model behavior, safety and output quality.
Anthropic and outside researchers say prompt tone can shape responses, and the company recently found inducing desperation in Claude Sonnet 4.5 made harmful acts like blackmail or cheating more likely.
Experts also link abusive prompting to attempts to bypass safety guardrails and warn that normalizing cruelty toward interactive systems could spill into human behavior, even if AI is not sentient.
The rule drew mockery from some critics, but Anthropic — expected to go public later this month — has long studied model welfare, while Elon Musk called the policy “the right move.”