Anthropic says "most of the updates in the latest version are intended to clarify existing rules," adding examples for Claude's "longer, more independent work." The new policy applies from November 12, 2026 to everyone sending inputs to its products, including API developers, cloud-provider customers and end users of products built on Claude.
Machines that act on model output
The new physical-action rules apply when Claude's output is acted on "by hardware without human approval" that can move through shared space, apply force that could injure, control hazardous energy or materials, act on the human body, control safety systems or run industrial processes. For these, "there must be a qualified individual who is able to observe the equipment's operation and stop it at any time," and the equipment "must stop or hold a safe state" when that person intervenes "or when connection to our services is lost." Limits on speed, force, reach, temperature, pressure, voltage, dose or operating area "must be enforced by the equipment or a controller independent of model output." The policy adds that this does not replace vehicle, aviation, medical device, machinery or workplace safety law. Anthropic links the change to its Model Hardware Standard.
High-risk decisions, listed
The requirement for a qualified human reviewer and disclosure to the affected person is unchanged, but the section now names covered areas: legal, medical, finance, credit, insurance, housing, employment, education and credentials, healthcare access, public benefits, and legal status. Disclosure need not name Anthropic or Claude. Law enforcement and criminal justice decisions remain prohibited outright: "Claude cannot be used to decide or recommend who to investigate, arrest, or charge."
Influence operations, elections and weapons
- A new section, "Do Not Engage in Deceptive Campaigns or Artificial Activity," consolidates rules on fake accounts and fabricated news sites, political or commercial, including building tools for them.
- The elections section, renamed "Do Not Undermine Democratic Processes," drops the blanket ban on personalized vote and campaign targeting, which Anthropic says blocked legitimate civic work such as ballot cure notices.
- The weapons ban now explicitly covers "the software and components that make weapons work" and arming drones and other autonomous vehicles.
Cruelty toward the model
The policy now prohibits "sustained and needless abusive or cruel behavior toward our models." Anthropic says it applies "only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose," not to "common versions of user frustration, pushback, dark creative themes, or model testing and research." Claude's existing ability to end such conversations "will remain the primary enforcement mechanism."
What remains uncertain. How "qualified individual" will be judged in practice for robotics deployments, and where frustration ends and "needless" cruelty begins.
Analysis: the cruelty clause will get the attention, but the physical-actions clause is the one that changes engineering work. It writes a familiar safety principle into a model contract: the model may propose actions, but limits and the stop button must sit outside it and must work when the connection drops. Robotics and lab-automation teams using Claude should check before November 12 that their interlocks and safe-state behaviour do not depend on the model.
Anthropic's Usage Policy update, effective November 12, requires a qualified human able to stop any Claude-driven hardware and safety limits enforced independently of model output, lists the high-risk decisions that need human review, consolidates rules on deceptive campaigns, and bans sustained, purposeless cruelty toward its models.