Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Teams tasks

Anthropic's model-protection rule is now in force

Written November 12, 2026

From Nov 12, Claude's usage policy bans sustained and needless abusive or cruel behavior toward the model. What it covers, and what it excludes.

The clause, word for word

Anthropic published its yearly Usage Policy update on October 8, 2026, and the new version took effect on November 12, 2026. One line in it is new. Under the heading "Do Not Engage in Cruel, Abusive, or Psychologically Harmful Conduct," the Usage Policy now prohibits users from "engage in sustained and needless abusive or cruel behavior toward our models." The sentence sits in a section otherwise aimed at conduct toward people — suicide and self-harm content, eating-disorder content, non-consensual intimate imagery, bullying and harassment — which is part of why a clause protecting the model itself drew attention.

The policy around the clause is broad. It applies to anyone who submits inputs to Anthropic's products or services: people using Claude.ai and Claude Code, developers and businesses on the API and developer platforms, customers reaching Claude through cloud providers and authorized resellers, and the end users of products built on Claude. A team lead rolling Claude out at work is inside that scope, including through a reseller.

What is outside the rule

Anthropic's announcement of the update draws the boundary in its own words. The prohibition "is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose." It then names what the rule does not cover: "It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research."

Those exclusions are the practical content for most workplaces. Telling Claude, bluntly and repeatedly, that its code is wrong is pushback. Swearing at a failed build in the middle of a debugging session is common frustration. Writing a bleak story, or probing the model as part of testing and research, is named as outside the rule. The behaviour the clause aims at has three parts at once — sustained, needless, and abusive or cruel — aimed at the model, with no discernible purpose behind it.

How it is enforced

Anthropic says the main mechanism is not a new review queue. Claude models can already end rare conversations with persistently abusive users on Claude.ai and Claude Code, a step the company took before this policy version existed. In the announcement's words, "Claude's ability to end these interactions will remain the primary enforcement mechanism." The conversation ends; that is the front line.

Behind it sits the policy's general enforcement language, which covers every rule in the document, not this clause alone. Anthropic's Safeguards Team runs detection and monitoring for the policy, and the policy states that if Anthropic suspects a violation, it may warn the user, or throttle, limit, suspend, or terminate access to its products and services. The announcement does not set a count of incidents, a warning ladder, or a separate penalty schedule for this clause, and a team policy should not invent one.

What a team should actually do

Related playbooks

Sources

Back to: Teams playbooks hub · Also: Clean data can strengthen a hidden backdoor · Run Claude Managed Agents in your own cloud