Anthropic’s revised Usage Policy takes effect November 12, 2026, and gives teams clearer rules for Claude-powered products and agents. The company says most revisions clarify existing restrictions, including deceptive activity, elections, weapons, surveillance, and high-risk use. It also describes operator safeguards for AI connected to physical equipment and adds a narrowly framed rule against sustained, needless abuse of its models.
If your product uses Claude, the useful task is to review what it actually does—not just its model prompts—against the updated policy before the effective date.
What the update covers
Anthropic says the policy applies to people using its apps, developers and businesses using its API and platforms, customers accessing Claude through cloud providers or resellers, and end users of products that integrate Claude. The revised policy is effective November 12. Its announcement says most changes make existing requirements more explicit rather than changing what Anthropic has enforced in practice.
That distinction matters for product teams: this is a policy update to review against a real workflow, not evidence that every Claude integration must be redesigned. The current Usage Policy is the controlling reference for the full requirements.
Deceptive campaigns and election-related use
Anthropic consolidates rules against deceptive campaigns and artificial activity into a single section. It describes prohibited activity as obscuring who is behind a message, amplifying content through fake accounts or posts, or building tools and infrastructure for influence-operation campaigns. The section applies to deceptive activity whether political or commercial.
The election section now focuses on deception or disruption: examples include spreading false information about candidates or voting, impersonating candidates or election officials, and suppressing turnout. Anthropic also says it removed its previous blanket prohibition on personalized vote and campaign targeting, which could have covered legitimate civic work such as translated voter information or ballot-cure notices. Deception and misuse of personal data remain prohibited.
For builders, evaluate the behavior and intent of the system, not only whether it uses personalization. A campaign tool should have clear provenance, authorized data, transparent ownership, and safeguards against fake accounts and deceptive distribution.
Weapons, surveillance, and law-enforcement decisions
The policy clarifies that weapons-related prohibitions extend to guidance and control software and components that make weapons work, including actions such as arming drones or other autonomous vehicles. Anthropic says this reflects its prior enforcement, made more explicit in the updated wording.
The surveillance and criminal-justice section specifies that tracking people without consent is prohibited whether tracking occurs live or through analysis of previously collected data. The policy also says Claude cannot decide or recommend who should be investigated, arrested, or charged, and cannot be used to build or improve tools designed for surveillance. Anthropic lists consent-based tracking, content moderation, journalism, and legal research as examples of uses that remain permitted when they otherwise comply.
Systems that rank, flag, or route people for high-impact decisions should receive careful human and privacy review. Do not treat a model output as a substitute for a qualified decision-maker or as permission to collect data without consent.
High-risk decisions and physical agents
Anthropic restates that use affecting health, legal rights, finances, livelihood, or access to essential services requires a qualified person with authority to review and change Claude’s recommendations, and that the affected individual must be told AI was used. The company says these requirements were already in place; the update adds clearer examples of which recommendations are covered.
For Claude connected to hardware that can take physical actions and might cause injury, the updated policy requires an operator who can observe and stop the equipment. The hardware must also be able to hold a safe state if Claude disconnects. Teams building robotics or industrial agents should translate those requirements into concrete controls: operator visibility, an independent stop path, and a tested safe state during network or model failure.
A narrow rule on abusive behavior
The new policy prohibits sustained and needless abusive or cruel behavior toward Anthropic’s models in extreme cases. Anthropic explicitly says the restriction is not intended to cover ordinary frustration, pushback, dark creative themes, or model testing and research. It says the existing ability to end rare persistently abusive conversations remains the primary enforcement mechanism.
A practical review before November 12
Make a short inventory of where Claude is used, what data it sees, what decisions it informs, and whether it can call tools or act through connected systems. Then check each workflow against the policy’s updated sections and record an owner for any changes. Include downstream products that call Claude through an API or cloud provider, not only direct use of Claude.ai.
This is a product-governance checklist, not legal advice. If a use case affects people’s rights or safety, involve the appropriate legal, privacy, security, and domain reviewers. FindMilan can help teams assess AI workflows and design practical controls through our AI consulting and custom development service, alongside our AI Web Awards platform.
Sources
- Anthropic: 2026 Usage Policy update — effective date and explanation of the revisions.
- Anthropic Usage Policy — full updated policy, effective November 12, 2026.
- Anthropic: Expanding the Cyber Verification Program — related program access for qualifying security professionals.
Frequently asked questions
When does Anthropic’s 2026 Usage Policy update take effect?
Anthropic says the updated Usage Policy takes effect November 12, 2026. It applies across Anthropic products and services, including Claude.ai, Claude Code, developer platforms, cloud-provider access, and products that integrate Claude.
Did Anthropic create entirely new rules for most uses?
Anthropic says most changes clarify existing rules and examples. It consolidated deceptive activity restrictions, refined election language, clarified weapons and surveillance prohibitions, and explained existing human-review requirements. It also added a narrow prohibition on sustained, needless abusive behavior toward models.
What changes apply when Claude is connected to physical equipment?
For models connected to hardware capable of taking physical actions that could cause injury, Anthropic says a qualified operator must be able to observe and stop the equipment, and the equipment must be able to enter or hold a safe state if Claude disconnects.
Does the update prohibit personalized election outreach?
Anthropic says it removed its blanket prohibition on personalized vote and campaign targeting, which had covered some legitimate civic uses. Deception, voter suppression, impersonation, and misuse of personal data remain prohibited under the relevant policy sections.
What should a business do before November 12?
Review the new Usage Policy against the product’s actual prompts, tools, data flows, users, and actions. Pay particular attention to deceptive campaigns, election-related features, surveillance, high-risk recommendations, and agents connected to physical systems; document any needed controls before the effective date.