Anthropic Rewrites Usage Policy, Adds Physical-Action Rules

Anthropic published a new version of its Usage Policy on 8 October 2026, taking effect on 12 November 2026. The company says most of the changes clarify existing rules, with new examples of how they apply as Claude takes on longer, more independent work.

The update adds requirements for when Claude controls hardware that takes autonomous physical actions, consolidates its rules on deceptive campaigns into one section, narrows its elections section, and prohibits sustained and needless abuse of its models.

Rules for Physical Actions

Under the new policy, when Claude’s outputs are acted on by hardware without human approval in high-risk settings, there must be a qualified individual able to observe the equipment’s operation and stop it at any time. The equipment must stop or hold a safe state when that person intervenes or when the connection to Anthropic’s services is lost.

Operating limits such as speed, force, temperature or dose must be enforced by the equipment or a controller independent of model output, the policy says. It lists six kinds of high-risk physical action, including moving through shared space, applying force that could cause injury and acting on the human body.

The existing rules for high-risk recommendations in areas such as legal, medical, finance, credit and employment have not changed, Anthropic says: a qualified human in the loop must review the recommendation, and the affected person must be told AI was used. The section has been rewritten to list which recommendations are covered and which are not.

Deception, Elections and Surveillance

A new section, Do Not Engage in Deceptive Campaigns or Artificial Activity, applies to political or commercial deception, including fake personas and outlets and building tools for influence operations. Anthropic says the restrictions were previously scattered across its elections, fraud, privacy and disinformation sections.

The elections section is renamed Do Not Undermine Democratic Processes and focused on deceiving voters or disrupting elections. Anthropic says it removed its blanket prohibition on personalized vote and campaign targeting because it covered legitimate civic work, such as election officials sending ballot cure notices.

The surveillance section now states that Claude cannot be used to decide or recommend who to investigate, arrest or charge, and that tracking people without consent is prohibited whether in real time or from previously collected data. The weapons rules now make clear they include the software and components that make weapons work, and actions like arming drones; Anthropic says both changes reflect how it already enforced the policy.

Abuse of the Model

The policy adds a prohibition on “sustained and needless abusive or cruel behavior toward our models.” Anthropic says it applies only to extreme cases with no discernible purpose, not to ordinary frustration, dark creative themes or model testing, and that Claude’s ability to end such conversations remains the primary enforcement mechanism.

Anthropic also clarified its Supported Regions Policy: use is prohibited by people physically located in an unsupported region, by entities incorporated or headquartered in one, and by entities majority-owned or controlled from those regions.

The Structural Read

Most of it is clarification. Anthropic says most updates clarify existing rules, and that its weapons and surveillance changes reflect how it already enforced the policy.

Anthropic names one rule it removed. It dropped its blanket ban on personalized vote and campaign targeting, saying it covered legitimate civic work.

The exclusions are spelled out. The policy says the physical-action requirements do not apply to commands that a qualified person reviews before they are executed on equipment.

Three Implications

EFFECTIVE 12 NOVEMBER The new policy takes effect on 12 November 2026, Anthropic says.

A STOP FOR HARDWARE A qualified individual must be able to stop high-risk equipment at any time, the policy says.

DECEPTION IN ONE SECTION Political and commercial deceptive campaigns are now covered by one section of the policy.

The Business Engineer Lens

This story maps onto the Business Engineer framework Inside Anthropic’s Permission Layer.

The framework’s starting point: Anthropic “is becoming the first major AI lab actively attempting to define the governance layer of the AI stack.”

As we read it, the Usage Policy is one of the documents where that layer is written down: it sets who may use Claude, for which purposes and under what conditions, and with this update those conditions extend to hardware that Claude controls.

What Is Not Established

We read Anthropic’s post and the full Usage Policy page; we did not compare it line by line with the previous version. The post does not say how many accounts the earlier rules affected. We did not contact Anthropic.

The Bottom Line

Anthropic’s Usage Policy, effective 12 November 2026, requires a qualified person able to stop hardware that Claude controls in high-risk settings, gathers its deception rules into one section, narrows the elections rules and prohibits sustained, needless abuse of its models.