What you can paste into AI — and what you really shouldn't
The career-saving rule, in one sentence.
By the end of this lesson
After this lesson, you'll have a one-sentence rule for what's safe to paste into a public AI — and a clear understanding of how that rule changes when your company gives you an enterprise tool.
This is the lesson nobody sat you down for. The one most quiet AI-related career stumbles trace back to.
The good news: it isn't complicated. There's a one-sentence rule that holds in most workplaces, and once you know how it changes between the public version of an AI and the version your company licensed, you're already ahead of most colleagues.
The one-sentence rule
Here's the rule that's saved a lot of careers.
If you'd email it to a stranger outside your company, you can paste it into a public AI. If you wouldn't, don't.
That's it. The rule is honest because it tracks the actual risk: does this leave my organisation's boundary? A public AI like the free version of ChatGPT or Gemini is, contractually, a stranger outside your company. The conversation may be retained, may be reviewed, and may be used to improve the model. The provider has a privacy policy, but it isn't your employer's privacy policy.
Apply the test honestly. Sales numbers you'd happily put on LinkedIn? Fine. The customer complaint with the customer's actual name still in it? No.
A Mumbai sales manager wants AI to redraft a customer apology. She pastes the email — but first she swaps the customer's real name for "Customer A" and removes the order number. Now it's just text she'd happily share with a stranger. Safe to paste.
The list, made specific
Some categories are nearly always safe in the public version. Some are nearly always not. The middle is where most quiet trouble happens.
Almost always safe to paste into a public AI:
- Your own writing — drafts, blog posts, internal notes you'd happily share.
- Public-facing copy you're editing — anything that's already on your website.
- Sample data with names, account numbers, and identifying details removed.
- Code that's already open-source.
- Templates, agendas, and structures.
Never paste into a public AI:
- Customer data — names, emails, account numbers, support tickets.
- Internal source code, especially anything containing API keys, passwords, or secrets.
- Pre-release financials, board papers, M&A documents, sealed bids.
- Anything covered by an NDA — yours or someone else's.
- Patient health information, legal advice from your company's lawyer, HR records (pay, performance reviews, complaints).
The middle ground — drafts that aren't public yet but aren't sensitive either — is judgement territory. The rule keeps holding: if you wouldn't email it to a stranger, you wouldn't paste it.
Samsung learned this publicly in 2023. Three separate internal incidents were widely reported in tech press: engineers pasted internal source code, internal meeting recordings, and confidential factory data into the public ChatGPT, looking for help with their work. The data left the company boundary. Samsung quickly restricted generative AI on company devices afterwards. Nothing suggests the engineers were trying to leak — they were trying to be useful. The rules clearly hadn't landed where they needed to.
Why enterprise AI changes the rules (mostly)
If your company has rolled out an enterprise AI — Microsoft 365 Copilot, ChatGPT Enterprise, Azure OpenAI, Google Gemini at Work, an internal Claude — that's a different contract. The brain is the same; the boundary is different.
Three things usually change with enterprise AI:
- Your data stays inside your company's tenant — its own private space inside the cloud provider. Often in a specific region (your country, the EU, the US). Easier to comply with privacy rules.
- The provider contractually agrees not to train on your prompts. The AI won't get smarter from your work and won't accidentally surface it to other customers later.
- Your IT team has audit logs — they can see who used what, set guardrails on sensitive content, and turn features off for specific groups.
This usually means the company tool can handle more work material than the public version — but customer data, contracts, source code, and HR material still depend on the rules your IT team has set.
The word "usually" matters. Enterprise AI isn't automatically safe. Three things to confirm before you treat it as inside-the-boundary:
- Has IT actually configured it for your team? (Some rollouts leave certain content blocked by default.)
- Are there categories your company specifically restricts? (Many do — patient data, legal privilege, source code with secrets.)
- Is it the company-sanctioned tool, or a personal account that just happens to share a brand name? (Microsoft 365 Copilot inside your work tenant is enterprise. Copilot.microsoft.com signed in with your personal account is public. Easy to mix up.)
If unsure: ask IT. The question "is it OK to paste customer data into Copilot?" is the kind of question IT teams would usually rather answer than clean up a bad guess later.
An Edinburgh financial analyst pastes a draft client report into her company's Microsoft 365 Copilot. The data stays inside her firm's tenant, isn't used to train the model, and IT can audit her usage if needed. She'd never paste the same draft into the free ChatGPT.com — same brain, different contract.
Try this in 5 minutes
Open your work email, your Slack/Teams, or whatever you used today. Pick three things you might have pasted into AI this week. For each one, sort it:
- Safe in the public version (you'd email it to a stranger)
- Safe in your enterprise AI but not the public version (it's company-internal but not super-sensitive)
- Not safe anywhere — this one needs human-only handling
If you're not sure which AI tool your company has sanctioned for your role, that's the next thing to find out. Ask your manager or IT before pasting anything from category 2.
Keep this
- Rule — If you'd email it to a stranger outside your company, you can paste it into a public AI. If you wouldn't, don't. Enterprise AI changes the rules — but only after IT has actually configured it.
- Phrase — "Same brain, different contract." — public ChatGPT and your work Copilot are not the same tool, even when they look identical.
- Don't — Don't assume an enterprise AI is automatically safe for everything. Don't paste customer data, source code, or HR records into a public AI as a workaround when your enterprise tool is slow or annoying. The career damage isn't worth the convenience.
Pop quiz, no marks
-
A colleague pastes a customer's full email — name, address, complaint — into ChatGPT.com to draft a reply. What's the friendly correction in two sentences?
Show answer
ChatGPT.com is governed by a consumer contract, not your employer's. The customer's data has now left your company's boundary — strip the name and details first, or use the company's sanctioned AI instead.
-
Your IT team has rolled out Microsoft 365 Copilot. Does that mean it's safe to paste anything into it?
Show answer
Mostly, but not automatically. Confirm the configuration covers your role, check whether your company restricts specific content categories (patient data, source code with secrets, legal privilege), and make sure you're signed in with your work account — not your personal one.
Want to go deeper?
Each of these topics on Plain AI explores one idea from this lesson in more detail: