The short version: Anthropic’s updated Claude usage policy, published October 8 and in force from November 12, 2026, now names “setting or changing shifts, hours, or pay” as a high-risk use. If Claude recommends who works Saturday and you post that schedule, two things are required: a qualified person reviews it with the power to change it, and the people on it are told AI was used. If you decide the schedule and Claude only writes it up, neither rule applies. And a line that used to cover any business publishing AI-written content has come off the list entirely.
Anthropic’s announcement says the high-risk requirements “haven’t changed” and that the section was rewritten so customers can tell more easily whether a use is covered. We read the new policy line by line against the version in force since September 15, 2025 (both read October 12, 2026). The rules themselves are the same. The list of what they cover is not, and the differences land on ordinary owner tasks.
What changed in the Claude usage policy?
The policy applies to anyone who types into Claude: people using the Claude app, businesses on the API, customers who reach Claude through a reseller, and the end users of other companies’ products built on Claude. It has three parts. Universal rules apply to everyone. High-risk requirements apply when Claude is used to make a recommendation about a person’s health, legal rights, money, livelihood or housing. Additional guidelines cover chatbots, agents and products for minors.
The high-risk part is where a small business is most likely to touch it. Here is how the old and new text compare for the areas owners actually work in.
| Area | 2025 policy | 2026 policy (from November 12) |
|---|---|---|
| Employment | Hiring tools, resume screening and “other employment determinations” | Screening and ranking candidates, hiring, promotion, setting or changing shifts, hours or pay, performance reviews, discipline and termination, each named |
| Housing | Eligibility for housing, including leases and home loans | Approving, denying or ranking a rental or purchase application; ending a tenancy; eviction or foreclosure |
| Taxes | Folded into “financial decisions” | Advising a person on what to report, deduct or claim, or preparing a return for them to sign |
| Publishing AI-written content | High-risk: automatically generating content and publishing it for outside readers | Not on the list |
| Who must be told | The individuals or consumers shown the output directly, at the start of each session | The person who gets the advice or is the subject of the decision; you do not have to name Claude or Anthropic |
Read that way, Anthropic’s “haven’t changed” holds for the two requirements, review and disclosure. What moved is the map. A shift schedule was arguably covered before under “other employment determinations.” Now nobody has to argue: it is named. And the publishing line, which read literally covered any blog or newsletter that a business generated and published automatically, is gone.
Where is the line between a draft and a decision?
This is the part of the new text worth reading twice. The requirements do not apply to “internal drafting, research, summarization, or analysis” that is not the final recommendation delivered. They do not apply when Claude is “carrying out a decision a person has already made,” or applying a fixed rule or formula with no judgment about the individual. And they do not apply to business operations that do not advise or decide about a specific person.
So the line does not run between using AI and not using it. It runs between Claude choosing something about a person and Claude carrying out what a person already chose.
Take a six-person roofing crew building next week’s schedule. There are three common ways to use Claude for it:
- You decide, Claude writes it up. You know who goes to which job; Claude turns your notes into a clean group text. Not covered.
- You set a fixed rule, Claude applies it. Seniority picks Saturdays first, nobody goes over 50 hours. If the rule leaves Claude no judgment, not covered.
- You ask Claude who should work the Saturday tear-off, based on speed and attendance. Covered twice over: it sets shifts, and it evaluates workers. You review it with the authority to change it, and the crew is told AI helped produce it.
The same split works elsewhere. A landlord with four units can have Claude summarize three rental applications; that is internal drafting. Asking Claude to rank them is a high-risk recommendation, and the applicants must be told AI was involved. A two-preparer tax office can have Claude explain how a home-office deduction works, which counts as general information. Telling a specific client which deduction to claim is covered, and the person who reviews it must have the training to judge it and hold a license where the law requires one.
What does the disclosure look like in practice?
One sentence. The policy says the person must be “clearly told” AI was used, and that you do not need to identify Anthropic, Claude or the model. On a crew schedule, “Drafted with AI help, reviewed by Maria” at the bottom of the text does the job.
Separate from the high-risk rules, any customer-facing chatbot or AI agent must say it is AI, at the start of each chat or in the product’s interface. If you run a Claude-based chat widget on your website, check its first line. It is the same courtesy Google built into its own calling agent, which opens every call by saying it is an AI.
What else got sharper for small business?
The new section on deceptive campaigns names fake reviews and bots that claim to be human outright, alongside sockpuppet accounts. Using Claude to write reviews of your own business was already against the policy; it is now spelled out, and it was already a federal problem under the FTC’s rule on fake reviews and testimonials.
One new line is worth knowing if a marketing vendor pitches you “AI visibility.” The policy now prohibits seeding the sources that search engines or AI systems draw answers from with content that misrepresents who wrote it, such as networks of sites posing as independent sources that all say the same thing. If a vendor’s plan for getting you cited by AI assistants is a stack of fake review sites, and the vendor builds it with Claude, the policy forbids it.
The good news is the bigger part of this update. Most of what a small business does with Claude, drafting quotes, answering email, summarizing a contract, writing the blog, sits plainly outside the high-risk list, and the policy now says so in writing.
What should an owner do before November 12?
Three habits cover nearly every case.
Decide first, then ask Claude to write it up. It keeps you outside the high-risk rules, and it is better management anyway: the judgment about your people stays with the person who knows them, and the AI gives you back the hour you used to spend typing.
When Claude ranks or recommends about a person, add the review and the sentence. Hiring, scheduling, pay, tenants, tax advice. You read it, you can change it, and the person is told.
Write it into your AI rules. If you already keep a short list of what staff may and may not use AI for, as in our three-line AI policy, add one line: AI may draft decisions about people, and a named person makes them. Our practical guide to AI for small business covers where to start if you have no list yet.
The policy itself notes that it does not replace local law, and that the stricter rule wins where the two differ. Nothing here is legal advice; it is what the document says.
Frequently Asked Questions
Does the Claude usage policy apply to the free Claude app?
Yes. It applies to anyone who submits inputs to Anthropic’s products, including people using the Claude app on any plan, businesses on the API, customers reaching Claude through a reseller, and the end users of other companies’ products built on Claude.
When does the new Claude usage policy take effect?
The updated policy was published on October 8, 2026, and takes effect on November 12, 2026. It replaces the version that had been in force since September 15, 2025.
Do I have to tell my employees I used AI to make the schedule?
Only when Claude makes the recommendation about who works, when, or for how much. If you decided the schedule and Claude only wrote it up, or Claude applied a fixed rule without using judgment, the disclosure requirement does not apply. When it does apply, one clear sentence is enough, and you do not have to name Claude.
Can I use Claude to write my business blog or newsletter?
Yes. Publishing AI-generated content is no longer on the high-risk list in the 2026 policy. The universal rules still apply, so the content cannot be disinformation, fake reviews, fake personas or material that hides who is behind it.
Where in your week would the line fall: is Claude drafting what you already decided, or is it starting to make the call?
