Anthropic Updates Claude Usage Policy

Anthropic has released an updated Usage Policy for its Claude models, effective November 12. The changes are driven by new AI capabilities, customer feedback, and instances of misuse. The updated policy includes new restrictions on misinformation, the use of AI in electoral processes, and clarifies requirements for using Claude in high-risk areas. According to the updated policy, Anthropic has prohibited the use of Claude for creating surveillance tools and operating weapons, including targeting systems. The company has also consolidated rules against disinformation campaigns into a separate section, covering both political and commercial campaigns. The rules prohibit concealing the authorship of messages and artificially spreading content using fake accounts.

Additionally, the updated rules define requirements for using Claude in high-risk areas such as healthcare and finance. In these cases, a qualified professional is required to verify the model's recommendations. Anthropic has also clarified the norms regarding the use of Claude in the context of elections, renaming the section to "Do Not Undermine Democratic Processes." These rules now focus on prohibiting the use of Claude to mislead voters. Examples include spreading false information about candidates or voting procedures, impersonating candidates or election officials, and attempting to impede voter turnout. However, the absolute ban on personalized voter targeting has been lifted, provided it does not involve deception or misuse of personal data.

Separately, the company has rewritten the rules on surveillance after noting an increase in the use of AI to create systems for identifying and tracking political dissidents. The updated policy prohibits tracking individuals without their consent—both in real-time and through the analysis of previously collected data. Claude cannot be used to make decisions about whom to investigate, detain, or charge in law enforcement processes. Creating and enhancing surveillance tools is prohibited. However, tracking with consent, content moderation, journalism, and legal research remain permitted. Finally, the policy includes a ban on prolonged and unjustified abusive behavior towards models. The rule will apply in exceptional cases where a user systematically abuses the model without a clear purpose. The ban does not extend to ordinary dissatisfaction, disputes, dark creative themes, testing, or model research.