Anthropic's Rewritten High-Stakes AI Rules: What Small Businesses Should Check by November 12

If you use Claude to rank job applicants, score rental applications or set a customer's credit limit, read this before November 12. That is when Anthropic's updated Usage Policy takes effect, spelling out when a qualified person must review Claude's recommendation before anyone acts on it, and when the person affected must be told AI was used. One line in the new policy may surprise you: setting staff shifts counts too. This is general information, not legal advice.

What Anthropic Changed

Anthropic's Usage Policy is the rulebook for what you may use Claude for. It covers the Claude app, businesses using its API (the connection developers use to plug Claude into other software), and the end users of products built on Claude. So if your hiring software runs on Claude behind the scenes, the rules reach you too.

On October 8, Anthropic announced that when its products "can affect someone's health, legal rights, finances, livelihood, or access to essential services," it requires a qualified "human in the loop" and "for the individual affected to be told that AI was used." Anthropic says these requirements "haven't changed." What it rewrote is the section explaining them, which "now lists which kinds of recommendations are covered and which are not." One wording change stands out. In 2025, Anthropic said the requirements applied "specifically when models’ outputs are consumer-facing, and not for business to business interactions." The new text says they apply "to anyone who uses our products to make a High-risk AI Recommendation," with no such carve-out.

Which Decisions Count

The policy names eleven "High-risk Areas." The ones small businesses touch most:

• Employment: "screening, ranking, advancing, or rejecting candidates," hiring, promotions, performance reviews, discipline, and "setting or changing shifts, hours, or pay."
• Housing: "approving, denying, or ranking a rental or purchase application," ending a tenancy or starting an eviction.
• Credit: approving, denying or setting the amount, rate or limit of a loan, credit line or similar product.
• Insurance, legal, medical and tax: pricing a policy or deciding a claim, advising someone on their own legal matter, diagnosing or treating, or preparing someone's tax return.

The rest cover investment advice, education and credentials, healthcare access, public benefits and legal status.

Does Your Use Count?

What a "Qualified Human" Means

Not just anyone clicking approve. The reviewer must "meaningfully review" the recommendation "with the authority to change or modify it," before it reaches the person or is used to implement a decision. A qualified person is "someone with the training or experience to evaluate the output in that field," licensed where the law requires it, and they stay responsible for the result. For a small landlord, that may simply be you.

The disclosure is simple. The person "must be clearly told that AI was used to produce it." You do not need to name Anthropic, Claude or the model.

What Happens If You Ignore It

If Anthropic suspects a violation, its policy says it "may warn you or throttle, limit, suspend, or terminate your access." Losing the tool your hiring or leasing runs on in a busy month is a real cost. The policy also says it does not replace local law. Colorado's new AI law adds its own notice duties (see our Colorado explainer), and California's newly signed AB 1609 bars large private businesses from presenting a customer chatbot as human, according to the law firm Covington.

Your Checklist Before November 12

1. List every place Claude touches a decision about a person, including bought tools that may run on Claude. Ask your vendors.
2. Name a reviewer for each, with real authority to override.
3. Reorder the steps so review happens before the decision goes out, not after.
4. Add one plain sentence to application pages, offers and rejections, such as "We use AI to help review applications; a person makes the final decision."
5. Check your customer chatbot. Anthropic separately requires consumer-facing chatbots built on Claude to tell users they are talking to AI, at the start of each chat or in the product itself.
6. Keep a simple log of who reviewed what, so you can show it later.

If step one turns up more than a couple of spots, that inventory is where every AI readiness assessment we run begins. Our small business AI policy template has review and disclosure sections you can adapt.

Sources

1. Anthropic, "2026 Usage Policy update" (October 8, 2026): November 12 start, human in the loop, disclosure, requirements unchanged. anthropic.com
2. Anthropic, "Usage Policy" (effective November 12, 2026): who it covers, High-risk Areas, review and disclosure rules, exclusions, enforcement, chatbot disclosure. anthropic.com
3. Internet Archive, snapshot of Anthropic's Usage Policy (October 16, 2026): the updated text as published. web.archive.org
4. Anthropic, "Usage policy update" (August 15, 2025): the earlier consumer-facing limit. anthropic.com
5. Littler, "Colorado Amends its Artificial Intelligence Law, Substantially Reducing Obligations on Employers" (May 15, 2026): Colorado's notice duty. littler.com
6. Inside Global Tech (Covington & Burling), "California Governor Signs Over 20 AI-related Bills into Law" (October 6, 2026): AB 1609 chatbot rules. insideglobaltech.com