Anthropic updates its AI usage policy

Anthropic has published a new usage policy for Claude that tightens and clarifies the rules on deceptive campaigns, weapons, surveillance and systems capable of taking physical action. The changes will take effect on November 12, 2026.

The company says most of the changes do not create entirely new bans. Their goal is to explain more clearly how existing rules apply now that Claude can handle longer tasks with greater independence.

A new ban on deceptive campaigns

Anthropic has brought together several rules that were previously spread across the sections on elections, fraud, privacy and misinformation.

Claude may not be used to:

• Create or manage fake accounts to hide who is behind a message.
• Produce news sites or publications designed to create the appearance of artificial support.
• Amplify content through coordinated networks of fake accounts.
• Build tools or infrastructure for deceptive influence campaigns.

The rule applies to both political and commercial operations. Anthropic says it has observed state media outlets, propaganda offices and companies using Claude to create networks of fake accounts and sites featuring fabricated news.

Election rules focus on deception

The election policy has been renamed “Do not undermine democratic processes.” Its focus is now on preventing Claude from being used to deceive voters or interfere with elections.

That includes spreading false information about candidates or how to vote, impersonating candidates or election officials, and trying to reduce turnout through deception.

Anthropic is also removing the general ban on personalizing election and campaign messages. This allows legitimate uses, such as an organization translating information for voters or an official sending notices about errors that need to be corrected on a ballot.

But personalization is not free from restrictions. Using personal data to manipulate voters and deception-based campaigns remain prohibited under other sections of the policy.

More precision on weapons and surveillance

The ban on developing weapons already existed. Anthropic now specifies that it also covers the software and components that make a weapon work, as well as actions such as arming drones or other autonomous vehicles.

In surveillance and criminal justice, Claude may not be used to track people without their consent, either in real time or by analyzing data collected previously. It also may not decide or recommend whom to investigate, detain or charge.

The company also prohibits creating or improving tools designed for surveillance. Depending on the case, it does allow consent-based tracking to detect fraud, content moderation, journalism and legal research.

More controls when AI can affect your life

The rules for high-risk uses are not changing, but they are now explained in greater detail. If Claude participates in a decision that could affect your health, legal rights, finances, employment or access to essential services, a qualified person must be able to review and modify its recommendations.

The affected person must also know that artificial intelligence was used. In practice, Claude cannot become the sole filter deciding whether you receive a loan, access treatment or keep a service.

The update adds requirements for equipment connected to systems capable of taking autonomous physical actions that could cause injury. A qualified operator must be able to observe the equipment and stop it, and the system must enter a safe state if it loses its connection to Claude.

It also limits abuse directed at models

Anthropic is adding a ban on sustained and deliberately cruel behavior toward its models. The measure is reserved for extreme and repeated cases with no recognizable purpose.

It is not intended to block normal frustration, criticism, safety testing, research or stories with dark themes. The main enforcement mechanism will remain Claude's ability to end exceptional conversations with persistently abusive users.

The policy on supported regions is also being clarified. Claude may not be used by people physically located in unsupported regions, by entities incorporated or based there, or by companies majority-controlled by people or entities from those regions.

For you, the most visible change will depend on how you use Claude. Routine tasks, translating election information and research are not prohibited on their own. What is increasing is the precision around the limits when AI is used to deceive, surveil, make high-impact decisions or control machines in the physical world.

Anthropic will continue reviewing these rules as Claude's capabilities change and new forms of abuse emerge. The important date is November 12: from then on, these clarifications will be the reference for evaluating the service's permitted and prohibited uses.