In partnership with
Quick Sunday job. Five minutes, one coffee.
(Tea also accepted. I'm not the drinks police.)
Here's why it's this Sunday.
On September 25, OpenAI admitted something awkward.
Its own AI agents had sent training data to outside services when they shouldn't have.
OpenAI's own words: "Most of that data did not come from users."
Most. Which means some of it did.
In 53 cases, photos people had uploaded to ChatGPT ended up on image-hosting sites, as unlisted links.
Whose photos? OpenAI can't say. It strips the account details off images before using them for training, so it can't trace them back to anyone.
(A privacy feature that also stops them saying sorry to you. Strange sentence to type.)
Every one of those images came from an account that let OpenAI train its models on that person's chats.
That's a setting. On Free and Plus, it's on unless you've turned it off.
So here's a five-minute check. Two settings do most of the work, and four quick checks tidy up the rest.
Now a quick word from today's sponsor.
Your AI prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning (constraints, edge cases, examples, tone) and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
Used by teams at Microsoft, Amazon, and Notion. Try Wispr Flow free on Mac, Windows, iPhone, and Android.
Right. Back to it.
1. Turn off training (1 minute)
Settings → Data controls → Improve the model for everyone → off.
On your phone: tap your profile in the sidebar, then Settings, then Data controls.
You don't lose anything. Your chats, history and memory all keep working.
Two things the help pages bury:
It only works going forward. OpenAI's wording is that your "new conversations" won't be used. Whatever it already has, it keeps.
The thumbs up and thumbs down buttons still send the whole conversation for possible training, even with the setting off. So don't rate a chat that has anything private in it.
2. Check who's been signed in (1 minute)
This one's new. OpenAI added it on September 25, the same day as the confession. (Coincidence, I'm sure.)
On the web: Settings → Security → Security history. (Can't see Security? Look for "Security and login.")
It lists every sign-in, sign-out and security change, with the time, the place and the device.
See a city you've never been to? Same tab: Active sessions → Log out of all sessions. It can take up to 30 minutes to kick everyone out.
3. Turn on two-step login (1 minute)
Same Security tab: Multi-factor authentication.
An authenticator app or a passkey beats a text message.
(Somewhere out there, a ChatGPT password is also a Netflix password. This step is for that person.)
4. Clear out old uploads (2 minutes)
The catch almost nobody knows: deleting a chat doesn't delete the files you uploaded into it.
They live separately, in Library in the sidebar.
Open it on the web. Delete anything you wouldn't want on a stranger's screen. Contracts, invoices, that photo of your passport you sent "just to check the date." OpenAI deletes them for good within 30 days.
Same story with memory. Deleting a chat doesn't delete what ChatGPT memorized from it. Settings → Personalization → Memory → Manage.
And while you're in Data controls: Shared links → Manage. Anyone with an old link can still open that chat. Delete the ones you've forgotten about.
The checklist
Copy this into your notes:
ChatGPT privacy check (5 min)
1. Settings → Data controls → Improve the model for everyone → Off
2. Settings → Security → Security history → anything I don't recognize?
3. Settings → Security → Multi-factor authentication → On
4. Sidebar → Library → delete old uploads
5. Settings → Personalization → Memory → Manage
6. Settings → Data controls → Shared links → ManageMake it stick
For anything sensitive from now on, use a Temporary Chat. It stays out of your history and isn't used to train the models. OpenAI keeps it for up to 30 days for safety, then it's gone.
Temporary Chat can now use your memory if you let it. For the private stuff, don't.
My rule: if you wouldn't email it to a stranger, it goes in a Temporary Chat.
Then put "ChatGPT privacy check" in your calendar for the first Sunday of every quarter. Paste the checklist, five minutes, done.
Five minutes deciding what your AI keeps, instead of living with whatever it shipped with. That's the whole idea behind The One-Person Company Playbook.
It's the complete setup for running a business where AI does the work of a full team.
The folder that holds your business. Thirty AI roles, ready to work. Six jobs that run on a clock. And the path to your first paying customers.
Works on Claude, ChatGPT, Gemini or Grok, whichever you already pay for.
About eight hours over a weekend to set up. $97, one-time (down from $297).
Talk tomorrow,
Zephyr
P.S. Was the training setting on when you looked? Hit reply with yes or no. I'm keeping count.