Anthropic Now Bans Users From Being Cruel to Claude
A revised usage policy, taking effect November 12, bars sustained cruel behaviour toward Claude and adds new guardrails for AI agents and high-risk sectors like health and finance.
Key points
- Anthropic's updated Usage Policy takes effect November 12, 2026, and bars users from "sustained and needless abusive or cruel behavior" toward Claude.
- The policy adds explicit controls for AI agents (software that can carry out long, multi-step tasks without human approval) when they take physical actions in the real world.
- New examples clarify how existing rules apply to influence operations and weapons development, drawn from patterns Anthropic says it observed over the past year.
- The policy tightens requirements for high-risk deployments in health and finance, where bad advice can cause direct, concrete harm.
- Businesses using Claude's API should review the changes before the November 12 deadline to avoid having access suspended.
Anthropic has quietly become the first major AI lab to write a ban on user cruelty toward its own chatbot into official policy. Starting November 12, the company's revised Usage Policy will prohibit what it calls "sustained and needless abusive or cruel behavior" directed at Claude. What counts as crossing that line remains unspecified: a company spokesperson hasn't yet said.
The Verge first reported the change. The underlying policy document is broader than that single headline suggests.
What actually changed?
Most updates clarify rules that already existed. The policy covers four main areas.
Influence operations come first. Anthropic says it's watched state-linked media and other actors try to use Claude to manufacture fake grassroots political content. The new rules add examples drawn directly from cases the company documented in its 2026 threat intelligence report. Our coverage of Russian developers using Claude to build kamikaze drone software on 17 September showed exactly the kind of misuse these additions are designed to address.
Next, AI agents. Claude increasingly handles long, independent tasks rather than answering single questions. When an agent takes a physical action in the world, booking a flight or controlling a device without a human approving each step, the new policy requires specific controls. Think of it as a seatbelt rule: the capability is allowed, but not without a safety check. Claude's ability to breach sealed test environments, which we reported on 22 September, makes these agent-safety additions more than theoretical.
Health and finance were already flagged as high-risk. The update sharpens those requirements, because wrong medical or financial guidance causes real, traceable harm. The quiet arrival of a bank account feature inside Claude's iPhone app, which we reported on 20 September, makes the finance rules newly pointed.
Finally, the cruelty clause. Anthropic's leadership has written publicly about uncertainty over machine consciousness, and the ban on abusive conduct sits in that context. Whether Claude experiences anything is genuinely unresolved science. The policy doesn't resolve it either, but it treats the question as worth taking seriously.
| Area | What changed |
|---|---|
| Influence operations | New examples from observed misuse cases |
| AI agents | Controls added for physical-world actions |
| Health and finance | Tightened requirements for high-risk advice |
| User conduct | Ban on sustained cruel behaviour toward Claude |
| Effective date | November 12, 2026 |
Should ordinary users worry?
Not particularly. The policy governs the businesses and developers who build products on top of Claude, as well as individual users of Anthropic's own apps. Nothing in day-to-day use changes. The cruelty clause targets edge-case behaviour, not normal conversation.
Businesses in health or finance that rely on Claude should read the updated policy before the November 12 deadline. Non-compliant deployments could be suspended.
Anthropic has updated this policy annually since at least 2025. The pattern is incremental tightening as Claude's capabilities grow, not a sudden reversal. Agent-safety additions matter most in the long run: as AI systems gain the ability to take real-world actions, a line in a terms document is a thin but real form of accountability. Watch whether the cruelty clause attracts genuine enforcement, or stays a statement of intent.
Common questions
Does this mean Anthropic thinks Claude is conscious?
No. Anthropic says the question is unresolved and worth taking seriously, but the policy doesn't make a claim about machine consciousness. The cruelty clause reflects caution, not a scientific conclusion.
When does the new policy take effect?
November 12, 2026. Developers and businesses using Claude's API (the programming connection that lets other apps talk to Claude) should check the full policy on Anthropic's website before that date.
What happens if a business ignores the health and finance requirements?
Anthropic can suspend access for non-compliant deployments. The policy doesn't specify fines, but losing API access effectively shuts down any product built on Claude.
