Anthropic has introduced a new policy that prohibits users from repeatedly engaging in abusive or cruel behavior toward its AI chatbot, Claude.
The updated Usage Policy, which takes effect on November 12, specifically targets sustained and unnecessary mistreatment of the AI. However, Anthropic clarified that ordinary frustration, criticism, creative writing, and legitimate testing will not violate the rules.
The company emphasized that the restriction applies only to extreme situations where users repeatedly abuse Claude without any meaningful purpose.
The decision follows research Anthropic began in August 2025, when it allowed Claude Opus 4 and 4.1 to end conversations involving persistent abuse. During testing, researchers observed signs of apparent distress in the models when users repeatedly pushed them toward harmful interactions.
Anthropic says ending abusive conversations remains its primary response, although it has not confirmed whether repeated violations could result in permanent account bans.
The policy update also introduces additional restrictions on coordinated fake online activity, the use of AI to arm drones, and potentially dangerous AI-controlled hardware.
The change highlights Anthropic's growing interest in AI welfare and raises questions about how people should interact with artificial intelligence, even as researchers continue debating whether AI systems can experience anything resembling human emotions.