Anthropic Just Changed the Rules for Apps Built With Claude. You Have Until 12 November.
TL;DR: Anthropic published a new Anthropic usage policy on 8 October, and it takes effect on 12 November 2026. The coverage led with the ban on being cruel to Claude. The part that actually lands on people who’ve built something is further down: if your app gives advice or drives decisions about money, health, housing, hiring, credit, insurance, education or legal status, you now need a qualified human reviewing those answers and you must tell the person AI was involved. You have about five weeks.
What actually changed
The new Anthropic usage policy was published on 8 October 2026. TechCrunch reported it the same day, under the headline the rest of the internet also picked: new rules against election interference, weapons software, surveillance — and a ban on sustained verbal abuse of the model.
That last one got the attention, and Anthropic was careful to limit it. The policy update, it said, “is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models”, and “does not apply to common versions of user frustration, pushback, dark creative themes”.
Fine. None of that is the part of the Anthropic usage policy that lands on you.
The part nobody reported: the anthropic usage policy’s high-risk list
Here’s what the coverage skipped. The policy itself carries an effective date — 12 November 2026 — and a list of what it calls High-risk Areas:
Legal · Medical · Finance · Credit · Insurance · Housing · Employment · Education and credentials · Healthcare access · Public benefits and services · Legal status and adjudication.
Those eleven areas are the whole of what the Anthropic usage policy treats as high risk. Everything outside them is unaffected.
If what you built produces advice or decisions in any of those areas, for other people, two obligations now attach to it. Neither is optional, and neither was in the headlines.
The two things you now have to do
The Anthropic usage policy spells both of them out in plain terms.
1. A qualified human has to look at it first. In the policy’s words, “a qualified person must meaningfully review a High-risk AI Recommendation, with the authority to change or modify it” — before the advice reaches the person, or before the decision is acted on. “Qualified” is doing real work there: that reviewer has to be qualified in the relevant field and hold any licence the activity legally requires. A general-purpose human clicking approve is not what this says.
2. You have to tell people AI was used. Anyone who gets advice, or who is on the receiving end of a decision based on it, “must be clearly told that AI was used to produce it”. You don’t have to name Anthropic, or Claude, or the model. You do have to say it.
Does this apply to you? Probably more than you think
The honest answer is that it applies to a lot of the things our readers build, because those categories are exactly where small, useful, obvious-looking tools cluster. The Anthropic usage policy wasn’t drafted with weekend projects in mind, but that is where a lot of them land.
Work through it:
| What you built | High-risk area | Caught? |
|---|---|---|
| A mortgage or affordability calculator that tells people what they can borrow | Finance, Credit | Yes |
| A tenant screening or referencing helper | Housing | Yes |
| A CV sifter or interview scorer | Employment | Yes |
| A “should I see a doctor about this” triage bot | Medical, Healthcare access | Yes |
| A tool that drafts or assesses contracts for clients | Legal | Yes |
| A benefits or grant eligibility checker | Public benefits and services | Yes |
| A course-marking or credential-assessment tool | Education and credentials | Yes |
| Your own internal pricing or margin calculator | — | No |
| A booking form, a landing page, a shop | — | No |
| A tool that only ever advises you about your business | — | No |
The dividing line the Anthropic usage policy draws is whether the output lands on another person’s money, health, housing, job, education or legal standing. Internal tools for your own business are untouched. So is anything that doesn’t give advice.
What else moved, briefly
Three other changes in the Anthropic usage policy are worth knowing, even though they’re unlikely to catch you:
- Autonomous hardware. If you connect Claude to equipment that takes physical actions on its own, a qualified operator must be able to watch and stop it, and the equipment has to hold a safe state if Claude disconnects.
- Surveillance. Tracking people without consent is out, live or from historical data, as is building tools to decide who gets investigated, arrested or charged. Consented monitoring, content moderation, journalism and legal research stay in.
- Supported regions. Access is now explicitly barred by where a person is physically located and where a company is incorporated, headquartered, or majority-owned from.
How to check your app this week
- Write down what your app actually outputs. Not what it does — what the user walks away believing or deciding. That’s the thing the anthropic usage policy is about.
- Check it against the list above. Money, health, housing, hiring, credit, insurance, education, legal status, public benefits. If it touches one of those for someone else, you’re in scope.
- If you’re in scope, decide who the qualified reviewer is. This is the expensive one, and it may be the moment you discover the product needs a professional attached to it. Better to find that out now than on 12 November.
- Add the disclosure. A plain line saying AI was used to produce this is enough, and it costs you nothing but an afternoon. Do it even if you decide you’re out of scope — it’s good practice and it’s increasingly the direction of travel everywhere.
- Write the date in your calendar. The new Anthropic usage policy bites on 12 November 2026.
Who should care (and who shouldn’t)
- Built something that advises other people about money, health, housing or work: this is aimed squarely at you. Steps 2 and 3 are this week’s job.
- Built an internal tool for your own business: you’re almost certainly fine. Read the list once to be sure, then get on with it.
- Built a website, a shop or a booking page: nothing here touches you.
- Still learning, nothing shipped: read the list anyway. It’s a surprisingly good map of which app ideas carry obligations and which don’t — and that’s worth knowing before you build, not after.
- Not using Claude at all: this specific Anthropic usage policy doesn’t bind you, but don’t file it as irrelevant. The direction is the same everywhere.
Our take
We think the disclosure rule is the one to adopt regardless.
Telling someone an answer came from AI costs you a sentence. It also happens to be the thing that most protects you if the answer turns out to be wrong — and in vibe coding, where the person who shipped the tool often can’t audit what it does — the same gap that left a lot of live apps relying on someone else spotting a framework flaw earlier this month — that’s not a small consideration. We’d add it to everything, in scope or not.
The qualified-reviewer rule is harder, and it deserves to be said plainly. For some of the app ideas people in this audience get excited about, it’s the difference between a weekend project and a business with a professional on the payroll. That’s not Anthropic being awkward. If your tool tells someone what they can borrow, or whether to see a doctor, a qualified person reading it before they act on it is simply what the responsible version of that product looks like.
Where we’d push back is on how this landed. The Anthropic usage policy governs every app built on the most popular coding assistant in this niche, it was published with a five-week clock, and the coverage was about whether you can swear at a chatbot. That’s not Anthropic’s fault — the policy page is clear and it’s public — but it does mean a lot of people who are in scope have no idea they are.
One practical note. Nothing here changes what you’ve already built or shuts anything off on 12 November. These are usage rules, not a technical switch. But the Anthropic usage policy is the agreement you’re building under, and the safest assumption is that it will be enforced the way the text reads, not the way the headlines did. Treat the five weeks as real — the same way we’d treat a spend cap you haven’t set yet.
Not sure whether what you want to build carries obligations like these? Take the 60-second Vibe Coding Tool Finder quiz →
FAQ
When does the new Anthropic usage policy take effect?
12 November 2026. Anthropic published the update on 8 October 2026, giving roughly five weeks’ notice. The policy page itself carries the effective date, so it’s worth reading the source rather than the coverage, which largely focused on the model-abuse and election-interference changes.
Do I have to tell users my app uses AI?
If your app produces advice or decisions in one of the high-risk areas — legal, medical, finance, credit, insurance, housing, employment, education and credentials, healthcare access, public benefits, or legal status — then yes. The policy says affected people must be clearly told AI was used. You don’t have to name Anthropic, Claude or the specific model.
Does the anthropic usage policy apply to internal tools?
Not in the same way. The high-risk requirements attach to advice and decisions that land on another person. A calculator you built to work out your own margins, a dashboard of your own numbers, or an automation that only ever affects your own business doesn’t trigger the qualified-reviewer or disclosure rules.
