Anthropic rewrites its Claude usage policy, adding a ban on sustained cruelty toward its models
This is the first usage policy overhaul in about a year. The abuse rule is narrow: it covers users who are repeatedly cruel to Claude with no apparent purpose, and explicitly excludes ordinary frustration, harsh feedback, dark fiction and model testing. Enforcement mainly relies on Claude's existing ability to end abusive conversations on Claude.ai and Claude Code, a feature that came out of the company's model welfare research. The election rules are consolidated into one section that bars false claims about candidates or voting, impersonating officials and voter suppression. Anthropic also dropped its blanket ban on personalized vote and campaign targeting, saying it blocked legitimate civic uses such as translated voter guides and ballot cure notices. Deceptive influence campaigns get a new consolidated section. Teams building on the API should review the new text before November 12, especially in civic tech, defense, surveillance-adjacent or finance and health products, where some prohibitions were clarified and others loosened.
