// AI Advantage
5 Minutes to Lock Down What AI Knows About You
The copy-paste Privacy Sweep Checklist — run it on every AI tool in your stack today.
Yesterday's Lesson made the point: opting out of training isn't true confidentiality, but it's still the fastest privacy win you've got, and most people have never flipped the switch. Worse, the defaults moved under you. GitHub Copilot started training on your code by default in 2026 unless you opted out (danilchenko.dev). Claude's October 2025 policy change quietly defaulted users who ignored the prompt into five-year data retention (mePrism). So today's edge is simple: spend five minutes, close the gaps, and stop leaking your work into someone else's training set. Here's the exact checklist.
Run these, top to bottom. Each is a real menu path verified this week.
PRIVACY SWEEP CHECKLIST — [DATE]
[ ] ChatGPT
Profile icon > Settings > Data Controls >
"Improve the model for everyone" > OFF
(Logged-out chats are trained on regardless — always log in.)
[ ] Claude
Settings > Privacy > "Help improve Claude" > OFF
Check this even if you set it before — the Oct 2025
default may have opted you into 5-year retention.
[ ] Google Gemini
myactivity.google.com/product/gemini >
"Gemini Apps Activity" > toggle OFF
[ ] Microsoft Copilot
Profile > [account name] > Privacy >
"Model Training on Text" > OFF
"Model Training on Voice" > OFF
[ ] GitHub Copilot
Settings > Copilot > Privacy >
"Allow GitHub to use my data for AI model training" > Disabled
[ ] THE REAL UPGRADE (for anything sensitive)
Move regulated/customer data off consumer tiers and onto
Team/Enterprise plans, which prohibit training on your
content BY CONTRACT — not just a toggle you can forget.
Sample output (what "done" looks like after five minutes):
ChatGPT ......... training OFF ✓
Claude .......... training OFF, retention 30 days ✓
Gemini .......... activity OFF ✓
Copilot ......... text + voice OFF ✓
GitHub Copilot .. disabled ✓
Sensitive data .. moved to Team tier (contractual) ✓
What changes: You stop trusting a default you never chose. Flipping these toggles tells the vendor not to reuse your chats for training. But here's the line that matters most: a toggle is a setting the vendor controls and can change — a Team or Enterprise contract is a promise you can enforce. Both ChatGPT Team (about $30/user/month) and Claude Team (about $25/user/month) prohibit training on customer content by contract (TrustScan). For anything regulated, roughly the price of a streaming bundle per seat is the difference between hoping and knowing.
Where else this works: Run the same sweep on your meeting transcriber, your CRM's AI add-on, your note app's "smart" features, and any browser AI extension. The AI bolted onto a tool you already trust often runs under different terms than the tool itself.
One hard truth before you go: a 2026 federal ruling held that AI conversations carry no legal confidentiality protection — there's no privilege for what you type into a chatbot the way there is with a lawyer or doctor (drainpipe.io). So these settings aren't a backstop. They're your front line, and they take five minutes.
One more move: forward this to the person on your team who pastes client data into ChatGPT without thinking twice. Five minutes of their time closes a hole in yours.
Settings protect what AI remembers about you. But where does your company's data actually live once AI touches it — and who has legal jurisdiction over it? That's tomorrow's Deep Dive on data sovereignty.
