Anthropic published an updated Usage Policy on October 8, 2026, taking effect November 12, that consolidates rules against deceptive campaigns, sharpens the prohibition on weapon development software, and adds new requirements for AI connected to hardware that takes autonomous physical actions. Most changes clarify existing rules rather than introduce entirely new restrictions, and they reflect the longer, more independent work the company’s models now perform.
Why the policy is being updated now
The 2026 revision responds to new patterns of misuse observed over the past year, including influence operations, weapons development, and surveillance of political dissidents. A separate threat intelligence report from Anthropic documented those patterns. The policy was also rewritten to give clearer guidance on high-risk use cases in health, finance, and law, and to address abusive behavior directed at the models themselves.
A new section on deceptive activity
Rules against using AI to run networks of fake accounts and fabricated news sites were previously scattered across sections covering elections, fraud, privacy, and disinformation. Those rules are now consolidated under a single heading: Do Not Engage in Deceptive Campaigns or Artificial Activity. The section applies to deceptive activity of any kind, whether political or commercial, and covers efforts to obscure who is behind a message, to amplify content through fake accounts, and to build the tools and infrastructure used to run influence campaigns.
Sharper rules on elections and democratic processes
The elections section has been renamed Do Not Undermine Democratic Processes and narrowed to focus on deceiving voters or disrupting elections. Examples covered by the new wording include spreading false information about candidates or voting procedures, impersonating candidates or election officials, and attempting to suppress turnout. A separate blanket prohibition on personalized vote and campaign targeting has been removed, because it had captured legitimate civic work such as nonprofits writing voter information in other languages and election officials sending ballot cure notices. Behavior that originally motivated that rule, including voter targeting based on deception or the misuse of personal data, remains prohibited under the deceptive campaigns, surveillance, and privacy sections.
Clarified weapons prohibitions
Use of the models to develop weapons has always been prohibited. The updated language makes clear that this prohibition also covers the software and components that make weapons work, and actions such as arming drones and other autonomous vehicles. According to the update, this language reflects how the rules have already been enforced.
More precise language on surveillance and law enforcement
The surveillance and criminal justice section has been rewritten to state more precisely what is prohibited. Tracking people without their consent is now explicitly banned, whether the tracking happens in real time or through analysis of previously collected data. The models cannot be used to decide or recommend who to investigate, arrest, or charge in a law enforcement or criminal justice process, and they cannot be used to build or improve tools designed for surveillance. Permitted uses, also spelled out in the new section, include tracking that people have agreed to (such as fraud monitoring), content moderation, journalism, and legal research.
Clearer requirements for high-risk use cases
When the products are used in ways that can affect someone’s health, legal rights, finances, livelihood, or access to essential services, a qualified human in the loop must have the authority to review and, if necessary, change the model’s recommendations, and the affected individual must be told that AI was used. These requirements are unchanged. The High-risk Use Case Requirements section has been rewritten to list more directly which kinds of recommendations are covered and which are not. Following the launch of a Model Hardware Standard, the section also adds requirements for use of the models when they are connected to hardware that takes autonomous physical actions and might be capable of causing injury. A qualified operator must be able to observe the equipment and stop it if needed, and the equipment must be able to hold a safe state if the model is disconnected.
Abusive behavior toward the models
The policy adds a prohibition on sustained and needless abusive or cruel behavior toward the models, intended to apply only in extreme cases where users repeatedly act cruelly with no discernible purpose. Common user frustration, pushback, dark creative themes, and model testing or research are not covered. The update aligns with a step already in place that allows the models to end rare conversations with persistently abusive users on the consumer and code products. Ending these interactions will remain the primary enforcement mechanism.
Clarified Supported Regions Policy
The Supported Regions page has been updated to clarify how restrictions are enforced. The policy prohibits use of the models by people physically located in an unsupported region, by entities incorporated or headquartered in one, and by entities majority-owned or controlled by persons or entities in those regions.
Looking ahead
Further revisions to the Usage Policy are expected as model capabilities and the risks they create continue to change. Input on future updates will be sought from inside the company and from policymakers, subject matter experts, civil society, and product users.
FAQ
When does the 2026 Usage Policy take effect?
The updated Usage Policy takes effect on November 12, 2026.
What new section was added to the Usage Policy?
A new section titled Do Not Engage in Deceptive Campaigns or Artificial Activity consolidates previous rules on fake accounts, fabricated news sites, and influence operations, applying them to deceptive activity of any kind, whether political or commercial.
What hardware safety requirements were introduced?
When the models are connected to hardware that takes autonomous physical actions and might be capable of causing injury, a qualified operator must be able to observe the equipment and stop it if needed, and the equipment must be able to hold a safe state if the model is disconnected.
This article summarizes reporting from anthropic.com.