Claude's New Rules Start November 12. Here Is the Part That Touches Your Business.
Anthropic rewrote its usage policy. Most of it is about weapons and elections, but a short list covers everyday work like hiring, rentals, taxes and patient triage.
By John Martines · October 14, 2026 · 6 min
Anthropic's updated usage policy takes effect November 12. If you use Claude to advise or decide about a specific person, a person has to review it and the customer has to be told. Here is how to check your own use in an afternoon.
A property manager with forty units has a stack of rental applications and a busy Friday. She pastes three of them into Claude and asks, "Which of these should I approve?" Claude ranks them and explains why. She approves the top one and sends the other two a polite no.
Nothing about that feels unusual. But under the rules Anthropic publishes for anyone using Claude, that ranking is a "high-risk" recommendation, and it comes with two conditions she probably didn't know about. A new version of those rules takes effect on November 12, 2026, and it spells out the conditions more plainly than before.
Every AI company publishes a usage policy: the list of things you may not do with its product, and the things you may do only with care. Anthropic, the company behind Claude, updates its policy once a year. The new version was published on October 8, 2026, and starts November 12.
Most of the headlines were about the unusual parts. The policy now bans "sustained and needless" cruelty toward Claude, which Anthropic says does not include ordinary frustration, pushback or dark fiction. It tightens rules on weapons, election deception and surveillance, and adds safety rules for machines Claude controls. Anthropic says most changes clarify rules it already enforced.
The part that matters to a small business is quieter. It is called "High-risk Use Case Requirements," and it covers using Claude to advise or decide about a specific person in eleven areas. Anthropic says the requirements themselves didn't change. What changed is that the policy now lists exactly what is covered and what isn't, so you can check your own use against it.
The eleven areas, in short: legal advice or documents, medical care and triage, personal investment or tax advice, credit, insurance, housing, employment, education and licensing, access to health care, public benefits, and legal status such as immigration.
When Claude's output is a recommendation about a person in one of those areas, two things are required:
- A qualified person reviews it. Someone with the knowledge and the authority to change the recommendation looks at it before it is used or delivered.
- The person affected is told. They must be clearly told that AI was used. You don't have to name Claude or Anthropic.
Some everyday work is specifically left out: general information that isn't about one person's situation, wellness tips, explaining a decision a professional already made, internal drafts and summaries that aren't the final recommendation, applying a fixed rule with no AI judgment, and business tasks that don't advise or decide about a specific person.
The policy applies to everyone who uses Claude, including people using it inside another company's product. If Anthropic finds a violation, it can warn, limit, suspend or end access.
How it got here:
- August 15, 2025: Anthropic published its previous update, effective September 15, 2025, adding rules for AI that acts on its own and for hacking.
- October 29, 2025: OpenAI moved to one usage policy across its products. It also names legal, medical, employment, housing, credit and insurance as sensitive areas where high-stakes decisions shouldn't be automated without a person reviewing them.
- October 8, 2026: Anthropic published the 2026 update.
- November 12, 2026: The new policy takes effect.
Most owners have never read an AI company's usage policy. Your staff haven't either. But these rules are part of the deal every time someone on your team uses Claude, on a free account or a company one, and they travel with any product built on Claude that you buy.
Three kinds of businesses should look closely.
Anyone who hires. "Employment" covers screening and hiring, shifts and pay, performance reviews, and discipline or firing. Asking Claude to rank applicants or draft a write-up that decides someone's fate is covered. Asking it to clean up a job posting is not.
Offices that advise people. Accounting and tax offices, insurance agencies, law offices, dental and medical practices, property managers, lenders. A tax answer for one client's return, a triage call on whether a patient needs to come in today, a coverage decision: each is advice about a specific person in a listed area.
Anyone with a chatbot on their website. The policy also says a chatbot or agent that talks with outside customers must tell them it is AI, at least at the start of each conversation or in the product itself. If your website chat runs on Claude, through your own build or a vendor's, that line applies.
A free account opens the rest of this article and the whole library.