Anthropic's updated Usage Policy takes effect November 12, 2026
TL;DR
Anthropic's updated Usage Policy takes effect on November 12, 2026. The headlines were about "be nice to Claude". The rules that matter for a business are three red lines, plus one privacy setting:
0Find your AI workflows. One audit prompt lists every task where AI writes, decides or talks to a customer, so you can run your whole audit in five minutes.
1Deciding something about a person? A qualified human reviews it before it reaches the person, with the power to change it. The person is told AI helped.
2Pretending to be someone? Say what it is. A character is fine. A con isn't.
3Watching someone who didn't agree? Consent, or it doesn't happen.
+The setting most people never checked. Settings → Privacy → Help Improve our AI models. Ten seconds, so you can keep your clients' details out of tomorrow's model.
Free: The Red Line Check cardThe full checklist on one printable page. One email unlocks it, plus the kits from every other video.
Everything from the video in one place, so you can copy it in seconds and fix your workflows before November 12.
0 The audit prompt: find your AI workflows
Paste into Claude:
The audit prompt
List every task in my business where AI writes something, decides something, or talks to a customer. For each one, tell me: does it decide something about a person, does it pretend to be someone, and does it track anyone without their consent?
So you can run your whole audit in five minutes.
✓ The Red Line Check: three questions for every AI workflow, before November 12
#
Question
If yes, do this
1
Does it decide something about a person? (money, credit, insurance, legal, medical, housing, jobs, education)
A qualified human reviews it before it reaches the person, with the power to change it. The person is told AI helped.
2
Does it pretend to be someone who doesn't exist? (fake reviews, fake accounts, fake personas, bots that claim to be human)
Say what it is. A character is fine. A con isn't.
3
Does it watch someone who didn't agree to it? (building a profile on a lead, an ex-client, a competitor's employee)
Consent, or it doesn't happen.
Examples from the video
Line 1: a bookkeeper's chatbot telling clients what they can deduct, a coach's bot handing out diet plans, a recruiter letting AI sort CVs and quietly bin half of them. "Qualified" means someone who could give that advice without AI: your bookkeeper, yes; your nephew who's good with computers, no.
Line 2: a website chat saying "Hi, I'm Sarah from support!" when Sarah is a bot. The fix: "Hi, I'm Sarah, an AI assistant." Sarah keeps her name; she just can't lie about being AI. AI-written isn't banned: writing your own LinkedIn post with Claude, as you, is fine. Inventing a customer who loves you is not. The line isn't who wrote it, it's who it claims to be.
Line 3: tracking with consent, like fraud alerts your customers signed up for, is fine.
1 The six words (for line 1)
Put this under any advice your AI drafts:
The six words
Drafted with AI, reviewed by [your name].
That way you keep using AI for the heavy lifting and stay on the right side of the line.
3 The consent sentence (for line 3)
If you track something, like who opens your emails, say it in plain words where people sign up:
The consent sentence
We track opens so we can send you less of what you ignore.
One sentence, so you keep the data and the trust.
! Bonus: the setting most people never checked
Who it applies to: the regular Claude plans (Free, Pro and Max).
If it's on: your chats can be used to train future models and kept for up to five years.
If it's off: the 30-day retention applies.
Where to find it: Settings → Privacy → Help Improve our AI models. It takes ten seconds, so you can keep your clients' details out of tomorrow's model.
Check it before you paste client data, not after. Anything already used for training can't be pulled back out.
Business plans and the API work differently. They're not used for training by default.
In the video this setting is called "Help improve Claude". Anthropic's help center now labels it "Help Improve our AI models" (Settings → Privacy). Same switch. Numbers re-checked against Anthropic's own pages on October 12, 2026.
★ For people who build AI tools for clients
Put the Red Line Check in your proposal, so you can win the clients who are nervous about AI and charge like the one adult in the room.
→ Tonight's to-do list
Run the audit prompt.
Fix the one workflow that fails.
Add the six words under anything your AI drafts for a client.
? What the headlines got wrong
"Anthropic bans weapons, mass surveillance and psyops." Those were already banned. The update spells them out with new examples. In Anthropic's own words, it doesn't change what they enforce in practice.
"You'll get banned for being mean." No. Normal rudeness is fine, and the usual result is a closed chat. Warn, throttle, suspend, terminate is the standard enforcement line for every rule in the policy. It isn't a new punishment for swearing at a chatbot. The rule targets sustained, needless cruelty (Anthropic's words: "no discernible purpose").
"Governments get special treatment." The clause exists, but it's been there since 2024, and the bans on disinformation, weapons and domestic surveillance still apply.
The one nobody mentioned: the update loosened something. The blanket ban on campaign targeting was dropped, because it was blocking nonprofits that translate voter information into other languages.
▶ Chapters
0:00 Cold open
0:42 The rule everyone's laughing at
2:53 What happens, and why
5:03 What the headlines got wrong
6:56 The Red Line Check
7:21 Red line 1: deciding something about a person
9:06 Red line 2: pretending to be someone
10:42 Red line 3: watching someone
12:01 Bonus: the setting most people never checked
13:07 The whole check
14:10 Outro
This is a plain-English summary, not legal advice. Read the policy itself: Anthropic's Usage Policy.
Victoria, the host in the video, is an AI avatar, and we say so. That's red line 2 in practice. The scripts, the voice and the strategies are mine.
Five real situations. No grades. It's a drill, so you spot a red line in your own business before November 12.
The part that counts
Used it in your business?
Watching is step one. Doing it is the whole point. When you've actually used this, tap the button, so your level shows what you've built, not just what you've watched.
Free resource
The Red Line Check card
The audit prompt, the 3 questions with their fixes, the six words, the consent sentence and the setting, on one printable page, so you can run the check on every workflow before November 12.
