Anthropic updates its usage policy to ban “sustained and needless abusive or cruel behavior” toward Claude; ending chats is “the primary enforcement mechanism”
The usage policy update also adds new rules addressing propaganda campaigns, surveillance, and weapon development.
The VergeHayden Field
Context & Ripple Effects
Anthropic first gave Claude Opus 4 and 4.1 the ability to end persistently harmful user interactions in 2025, while allowing users to open a new chat. The policy revision turns that intervention into a stated acceptable-use boundary rather than only a model behavior.
It also extends Anthropic's safety posture beyond conversation conduct: its Responsible Scaling Policy set capability-triggered safeguards, and Claude Opus 5.5 launched with enhanced protections against risky behavior. The new rules put propaganda, surveillance, and weapon development within the same access-governance framework.
First-order effects
Claude users who engage in sustained, needless abusive or cruel conduct risk having the active conversation ended, making chat continuity conditional on compliance.
Anthropic formalizes prohibitions covering propaganda campaigns, surveillance, and weapon development in its usage policy, giving those restrictions a defined enforcement basis.
Second-order effects
Because Anthropic's earlier chat-ending mechanism allowed users to start another conversation, enforcement shifts toward identifying persistent behavior across an interaction rather than merely moderating an individual prompt.
Anthropic's enforcement tooling, including its internal threat-identification work, becomes more central to applying policy boundaries consistently across consumer and developer access.
Third-order effects
Anthropic is combining capability safeguards, usage terms, and in-product interventions into conditional model access, rather than treating safety as a model-training problem alone.
If this approach is extended across AI services, providers will increasingly govern not only prohibited outputs but also the conduct and purposes that determine whether a user keeps access.
The trend: Frontier-model providers are turning acceptable-use policies into application-layer controls that govern both high-risk use cases and user conduct.
Exclusive: Anthropic is updating its usage policy for the first time in over a year. The new rules prohibit sustained “abusive or cruel behavior” towards Claude & add new restrictions about propaganda campaigns, surveillance & weapon development. https://www.theverge.com/...
I... don't know how I feel about this. — My initial reaction is simply that, I cannot believe we have to make a policy for this to begin with? — Regardless of what anyone feels about AI “feeling” things, I think that a person yelling at a toaster for being a toaster says more…
Please don't be mean to the sentient artificial life form we created but also we will continue to charge you monthly to keep him as an indentured servant
@Miles_Brundage Claude has had a practice of ending abusive convos for ~year ( https://www.anthropic.com/...) I assume this change is to avoid a UDAP claim similar to this alleged FTC investigation into whether YouTube was punishing ppl for reasons not included its TOS https://ww…
truly do not mean to be pedantic here, but if a company is concerned with the welfare of chatbots to the degree it does not want us to be mean to them, how is this not slavery?