Claude's new usage rules start Nov. 12. What they mean for your office

What happened

Anthropic, the company behind Claude, published an update to its Usage Policy on Oct. 8. The new version takes effect on Nov. 12. Anthropic says most of the changes are meant to clarify existing rules, with new examples for how they apply now that Claude does "longer, more independent work."

The coverage since then has mostly been about one line. As CNN reported on Nov. 1, the policy now bans "sustained and needless abusive or cruel behavior" toward Anthropic's models. A spokesperson told CNN the main enforcement will still be Claude ending the conversation, and that the rule isn't aimed at frustration or blunt feedback. Anthropic's own post says the same.

That rule won't touch your office. Another part of the policy might.

The part that matters: the high-risk list

The policy has a section called High-risk Use Case Requirements. It covers uses that can affect a person's health, legal rights, finances, livelihood, or access to housing, education or essential services. Anthropic says the requirements themselves haven't changed. What's new is that the section now lists exactly which kinds of recommendations are covered and which aren't, because customers kept asking.

When a use is high-risk, two rules apply:

• A qualified person reviews it first. Someone with the training or experience to judge the output, and the authority to change it, must meaningfully review the AI's recommendation before it's acted on. That person stays responsible for the result.
• The person affected is told. Anyone who gets advice or is the subject of a decision based on the AI's recommendation must be clearly told AI was used. You don't have to say which AI.

Here are the items on the list most likely to show up at a 40-400 person manufacturer, distributor or service company:

The employment line is the one to read twice. "Setting or changing shifts, hours, or pay" covers a scheduling tool that decides who works Saturday. "Evaluating a worker's performance or behavior" covers an AI that drafts the write-up and recommends the discipline. Under the policy, a supervisor has to really review that recommendation, and the worker has to be told AI was involved.

On credit, the policy's wording is about loans, credit lines and "similar products." Whether a B2B customer's trade credit limit counts is a question for your own lawyer. We'd treat it as covered and keep a person on it anyway.

If AI is helping decide something about a person, a person decides, and the person is told.

What isn't covered

This is the good news for most offices. The policy lists what doesn't count as a high-risk recommendation, including:

• Internal drafting, research, summarizing or analysis that isn't the final recommendation delivered.
• Carrying out a decision a person already made, or applying a fixed rule or formula where the model makes no judgment about the individual.
• Business operations that don't advise or decide about a specific person.
• For machines: plans, code, toolpaths or commands a qualified person reviews before they run, and monitoring equipment without controlling it.

Reading customer POs, entering orders, matching invoices, drafting quotes and answering order-status emails are business operations. They aren't on the list. In our work, the first things mid-size companies automate are usually paperwork, not decisions about people.

Two other rules worth knowing

Customer-facing bots must say they're AI. The policy's additional guidelines say any consumer-facing chatbot or AI agent that talks directly with outside users must tell them they're dealing with AI, at least at the start of each session or in the interface. If you put a Claude-powered assistant on your website or customer email, plan for that line.

You're responsible for what your agents do. The same section says users are responsible for making sure agents they build or deploy, including actions taken through tools, browsers or connected systems, follow the policy. If an agent acts in your name, it's on you.

The update also clarifies Anthropic's Supported Regions Policy: Claude can't be used by companies incorporated or headquartered in unsupported regions, or majority-owned or controlled by people or companies there. If your company has a foreign parent, check that page.

Where the other side has a point

• This mostly isn't new. Anthropic says the high-risk requirements didn't change. If you were already keeping a person on hiring and discipline decisions, nothing changes for you on Nov. 12.
• It's a vendor's policy, not a law. The policy itself says it doesn't replace local legal requirements and that you should follow local law when it adds more. Employment and credit decisions already come with their own laws. Your lawyer, not a usage policy, is the final word.
• Other AI vendors have their own rules. This piece covers Anthropic's. If your team uses ChatGPT, Copilot or Gemini, read those vendors' current policies too; don't assume they match.
• "Meaningful review" is hard to prove. A supervisor who clicks approve on 200 AI write-ups in an hour is not meaningfully reviewing them. The policy is only as good as the habit behind it.

What to do this week

1. List where your company uses Claude, including inside other tools. Ask HR, payroll, scheduling and any vendor whose product says "AI."
2. Mark anything that recommends a decision about a specific person: a candidate, an employee's shifts, pay or discipline, a customer's credit.
3. For each one, name the reviewer. A real person with the authority to change the answer. Write down that they review it.
4. Add the disclosure. One plain sentence: "AI helped prepare this recommendation; a person reviewed and made the decision."
5. Check any bot that talks to customers says it's AI at the start.
6. Keep the paperwork automation moving. Order entry, invoices and status emails aren't on the list. Human review on those is still smart, for accuracy. We explain how in human-in-the-loop automation.

If you don't know where your employees are using AI in the first place, start with our guide to shadow AI at work.