Anthropic has published an updated Use Policy for its Claude models; the new version will take effect on November 12. The company introduced additional bans on creating surveillance tools and enabling weapons, consolidated rules against misinformation campaigns, and clarified requirements for using AI in high-risk scenarios.
Unified rules against misleading campaigns
Restrictions related to misinformation and artificial activity are now grouped in a separate section, "Do not participate in misleading campaigns or artificial activity." This covers both political and commercial cases and prohibits concealing the sponsors of messages, artificially spreading content through fake accounts and posts, as well as creating tools and infrastructure for influence operations.
Changes on elections: lifting the total ban on targeting
The section on elections has been renamed "Do not undermine democratic processes." It is prohibited to mislead voters or disrupt elections, including spreading false information, impersonating candidates or officials, and attempting to interfere with turnout. However, the absolute ban on personalized voter targeting has been lifted: legitimate informational materials are allowed, while deception and misuse of personal data remain prohibited.
Bans on weapons development and software components
The policy now explicitly covers software and components that enable weapons, including guidance and control systems, as well as armament for drones and other autonomous vehicles.
Restrictions on surveillance and criminal justice decisions
Tracking people without their consent is prohibited, both in real time and using previously collected data. Claude cannot be used to make decisions or recommendations about whom to investigate, detain, or charge. Permitted uses include, in particular, consent-based tracking for fraud detection, content moderation, journalism, and legal research.
Requirements for high-risk scenarios
In areas where model recommendations may affect health, legal rights, finances, livelihoods, or access to essential services, qualified human oversight with the ability to adjust model recommendations is required.
