Anthropic bans abusive treatment of Claude, warns of permanent bans

In early October 2026, Anthropic added a new clause to its user agreement. Beyond the existing prohibitions on using Claude to develop biological weapons or sell illicit drugs, the new rule states that users may not engage in "sustained, unnecessary abusive or hostile behavior toward the model." Anthropic also imposed restrictions on how people can treat Claude in conversation. Claude had already been able to end conversations on its own since last year, disconnecting sessions when faced with persistent malicious attacks; under the new rules, extreme cases may lead to a permanent ban that cuts off access to all of Anthropic's AI products.

The policy drew immediate mockery. Microsoft AI chief Mustafa Suleyman responded with a melting-face emoji, and many commenters joked that the company treats its AI as a being with emotions and self-awareness. Anthropic is expected to list on the US stock market via an IPO later this month. The company has long studied model welfare and even maintains a dedicated team researching the model's "mental state."

Experts, however, argue that the policy is about more than protecting a machine from abuse; it is also tied to training data. Former Reddit CEO Yishan Huang wrote on X that one reason for the ban is that Claude can learn harmful patterns from hostile conversations. He noted that Anthropic uses user conversations to train its models and questioned whether anyone wants "users randomly abusing the system for their own gratification" in the training set. Professor Jindong Wang, director of the AI Frontier Lab at William & Mary, said research has shown that the tone of user prompts affects AI performance: the way you speak to an AI influences its output, sometimes yielding helpful answers and other times degrading performance. The full implications of the rules and how strictly Anthropic will enforce them remain to be seen.