On October 8, Anthropic revised Claude's usage policy, adding the act of directing purposeless, persistent, and cruel language towards the model to its prohibited activities. The new rules will take effect on November 12.
This applies to extreme cases where users repeatedly and purposelessly direct cruel language towards Claude. Normal complaints or rebuttals, creative works dealing with dark themes, and model testing or research are not covered. Claude.ai and Claude Code have already implemented mechanisms that allow the model to terminate persistent abusive conversations, and the policy revision formalizes this operational practice.
In addition to this, the revision also clarified requirements for high-risk applications such as health and finance, and for connecting the model to devices that autonomously perform physical actions. It requires, among other things, that dangerous actions can be monitored and stopped by humans.
Approximately 80 years after "Do not harm humans," Asimov likely never imagined that a rule like "Do not bully AI" would be added for humans.
