Anthropic updates Usage Policy: new Claude rules take effect November 12

Anthropic published a new version of its Usage Policy on October 8, 2026. The rules take effect on November 12, which leaves customers 35 days to review their products. The company revisits the policy once a year, and its explanation for this revision is simple: over the past year Claude has started handling longer, more autonomous tasks, and the rules need to show how they apply to these new capabilities. Most of the changes are clarifications. There is one exception: Anthropic has dropped its blanket ban on personalized targeting of voters and campaigns.

Fake accounts and disinformation now share one section

The change was prompted by the company's own observations. Anthropic has seen state media, government propaganda offices and commercial firms use Claude to run networks of fake accounts and counterfeit news sites. All of this was already formally banned, but the restrictions were scattered across the sections on elections, fraud, privacy and disinformation.

They are now consolidated in a new section, Do Not Engage in Deceptive Campaigns or Artificial Activity. It covers any deceptive activity, political or commercial: hiding who is behind a message, inflating content through fake accounts and posts, and building tools and infrastructure for influence operations.

The elections section has been narrowed and renamed Do Not Undermine Democratic Processes. It targets deceiving voters and disrupting elections: false information about candidates and voting procedures, impersonating candidates or officials, and suppressing turnout. The blanket ban on personalized targeting was removed because it got in the way of legitimate civic work. For example, nonprofits could not write voter materials in other languages, and election commissions could not send notices about ballot corrections. Targeting based on deception or misuse of personal data remains prohibited through the sections on deceptive campaigns, surveillance and privacy.

Weapons and surveillance: more specifics, same practice

The weapons section now states explicitly that the ban covers the software and components that make weapons work, including guidance and control systems. It also covers actions such as weaponizing drones and other autonomous vehicles. Anthropic has seen attempts to build such software with its models. The company stresses that the wording reflects practice it already applied.

The section on surveillance and criminal justice was rewritten after the September cyber threat report, which describes growing use of AI in systems that identify and track political opponents. Here too, Anthropic speaks of greater clarity rather than a change in practice.

• Surveillance of people without their consent is prohibited, whether in real time or using previously collected data.
• Claude cannot be used to decide or recommend whom to investigate, arrest or charge.
• Building or improving tools designed for surveillance is prohibited.
• Tracking with people's consent (for example, anti-fraud monitoring), content moderation, journalism and legal research are allowed.

What to check before November 12

The requirements for high-risk scenarios have not changed, but the wording is clearer. If a product affects health, legal rights, finances, source of income or access to basic services, it needs a qualified human in the loop who has the authority to review and, if necessary, change Claude's recommendation. The affected person must be notified that AI was used. The update also adds oversight measures for cases where Claude takes physical actions on its own, and provisions on abusive treatment of models. The source text of the announcement breaks off at this point, so for details see the full policy and the Supported Regions Policy.

In practice, teams building products on Claude should go through three items. First, check their automation against the deceptive campaigns section, especially marketing and account management. Second, make sure that medical, financial and HR scenarios have a real human reviewer and a notice for the user. Third, if they work in robotics, drones, security or the public sector, reread the sections on weapons and surveillance: the boundaries there are now drawn explicitly.