Anthropic adds a rule against repeated abuse of Claude

Anthropic has announced a restriction on prolonged, purposeless abuse directed at its AI models. The company published the revised policy on October 8 and said it would take effect November 12.

The company said the rule addresses exceptional cases involving repeated cruelty without an identifiable purpose. Everyday frustration and disagreement are outside its scope. It also exempts creative work involving dark subjects, along with research and model testing.

Claude can already stop certain abusive exchanges in Claude.ai and Claude Code, Anthropic said. The new rule puts that approach into its formal usage restrictions. Ending those conversations will remain the principal enforcement tool, the company said.

The change accompanies revisions covering deceptive activity, election interference and weapons-related software. Anthropic said it would continue revising its rules as the technology develops, drawing on input from users and outside specialists.