Anthropic bans sustained cruel abuse of Claude in policy update

Anthropic says it updates its usage policy as needed based on model evolution and user feedback, and describes this update as mostly a clarification of existing rules. The abuse clause is the genuinely new piece, and Anthropic ties it to a practice it says already happens: Claude sometimes faces persistent abusive behavior on Claude.ai and Claude Code, where Anthropic has taken steps to end conversations. The update is framed as bringing the written policy in line with that approach, which suggests the rule is codifying an existing response rather than introducing a new enforcement mechanism.

The rest of the update is organizational. Deceptive conduct that had been spread across several sections, including elections and fraud, is now grouped under a single heading on deceptive campaigns and artificial activity. Restrictions on political content were made more explicit. On weapons, armed drones and other autonomous vehicles were added to a prohibition that already existed, and surveillance-system restrictions were clarified.