Anthropic is expanding and reorganizing the rules governing Claude as its models take on longer, more independent tasks. TechCrunch reported on October 8 that the company’s updated Usage Policy adds or clarifies restrictions covering election interference, weapons software, surveillance, deceptive campaigns, autonomous hardware, and sustained abuse of the model. Anthropic says the revised policy takes effect November 12.
The company frames most of the changes as clarification rather than a new enforcement direction. In its own summary of the update, Anthropic said it drew on misuse patterns observed during the past year, including influence operations, weapons development, and surveillance. The result is a policy structured around the kinds of actions increasingly capable AI systems can help users perform, not merely the content they can generate.
A new section consolidates restrictions on deceptive campaigns and artificial activity. Anthropic says it has observed state media organizations, government propaganda offices, and commercial firms using Claude to help operate fake accounts and fabricated news sites. The revised rules prohibit efforts to conceal who is behind a message, artificially amplify material through fake accounts or posts, or build infrastructure for influence operations.
The election rules have also been narrowed and made more specific. Claude may not be used to deceive voters, impersonate candidates or election officials, spread false voting information, disrupt elections, or suppress turnout. At the same time, Anthropic is removing its blanket prohibition on personalized voter and campaign targeting because that rule also blocked legitimate civic work, such as translating voter information or preparing ballot-cure notices. Deceptive targeting and misuse of personal data remain prohibited.
Anthropic’s weapons language now explicitly covers the software and components that make weapons function, including efforts to arm drones or other autonomous vehicles. The surveillance section similarly states that Claude cannot be used to track people without consent, recommend who should be investigated, arrested, or charged, or build and improve surveillance tools. Anthropic says consensual fraud monitoring, content moderation, journalism, and legal research remain permitted.
The update also addresses physical systems controlled by AI. When Claude is connected to hardware capable of causing injury, Anthropic requires a qualified operator who can observe the equipment and stop it if necessary. The equipment must also be able to remain in a safe state if its connection to Claude is lost.
For high-risk decisions affecting health, legal rights, finances, employment, or access to essential services, Anthropic continues to require a qualified human who can review and change the model’s recommendation. The affected person must also be told that AI was used. The company says those safeguards are not new, but the policy now explains more directly which recommendations they cover.
The provision drawing the most attention may be the ban on sustained, needless cruelty toward Anthropic’s models. TechCrunch reported that the rule is intended for extreme, repeated behavior with no clear purpose, not ordinary frustration, criticism, dark creative material, or legitimate model testing. Claude can already end rare conversations with persistently abusive users, and Anthropic says ending the interaction will remain the main enforcement mechanism. The unusual clause may dominate discussion, but the broader update shows how AI policy is shifting toward agents that can influence institutions, control software, and act in the physical world.
Comments
Loading comments…