It's the most practical AI question an employee can ask, and most companies can't answer it.
The short version: anything you'd be fine posting in public and never getting back is safe to paste. Already-published material, generic questions, text with nothing sensitive in it. What's not safe is the rest of your actual work, the customer records, the source code, the contract, the financials, the unreleased plan, unless your organization runs a paid enterprise plan that contractually turns off model training and someone has decided that specific kind of data is allowed in.
That last clause is where almost everyone falls down. Most companies have never made that decision. So the real answer to "what's safe to paste?" is usually: nobody here has said, and you're guessing.
"Is ChatGPT safe?" is the wrong question
People treat safety as a property of the tool. They ask whether ChatGPT is secure, whether the vendor is trustworthy, whether the enterprise tier is locked down. Reasonable questions. They're just aimed at the wrong target.
Safety isn't a property of the tool. It's a property of the data you put in and where that data ends up. The same chatbot is completely fine for rewriting a job posting and a serious problem for debugging code that runs your product. Nothing about the tool changed between those two prompts. The data did.
So the better question is a simpler one: what am I about to put into it, and has anyone here decided that's okay? For most teams, the honest answer to the second half is no.
What actually happens when you paste
When you paste text into a consumer AI tool, that text leaves your company. On the free and personal tiers, it can be retained and used to improve the model. You can't pull it back. There's no undo.
Samsung learned this in the open. Within about three weeks of allowing ChatGPT internally, engineers had pasted proprietary source code and an internal meeting transcript into it on three separate occasions. Samsung banned generative AI tools on company devices shortly after. The exposure had already happened. A ban can stop the next paste. It does nothing about the one that already left.
This is not a rare event. Cyberhaven, analyzing usage across 1.6 million workers, found that 11% of what employees paste into ChatGPT is confidential, with internal-only data, source code, and client data topping the list. And most of it happens where no one is looking: in LayerX's 2025 report, 77% of employees paste data into AI tools, and 82% of those pastes come from personal accounts the company can't see. The pasting is constant, useful, and almost entirely invisible.
Why "use your judgment" isn't an answer
Faced with this, a lot of leaders land on "we trust our people to use good judgment." It sounds reasonable. It quietly pushes the hardest decision in the company onto whoever is most rushed.
Judgment needs something to judge against. An account manager on a deadline, trying to summarize a messy client contract, is not going to stop and reason out data-residency and model-training policy. They're going to paste the contract, get their summary, and move on, because no one ever told them that contract was different from the job posting. They aren't being reckless. They're filling a gap the company left open.
"Use your judgment" isn't a policy. It's the absence of one, with the risk transferred to the person least equipped to carry it.
So what's actually safe to paste?
The test that holds up is one question: if this exact text left the company permanently and showed up somewhere you didn't control, would that be fine?
If yes, paste it. If no, it doesn't go into any tool that hasn't been cleared for that kind of data. And if you genuinely can't tell, that's the most important result, because it means the line was never drawn for you.
Notice what that test needs to actually work: a shared understanding of which data is which. Public versus internal versus genuinely sensitive. That's data classification, and it's the quiet foundation under every confident answer to "what's safe to paste." Companies that can answer the question fast all have one. Companies that argue about it in a meeting don't.
The point isn't that AI is dangerous and your team should stop. It's the opposite. Once people know what's safe to put in, they can use these tools hard and without hesitating. The hesitation, and the risk, both come from the same place: a decision no one has made yet.
Find out what your team is already pasting
You can't classify data you can't see, and you can't decide what's allowed if you don't know what's already in use. The first step isn't a rule at all. Start with a clear read on where AI is already running in your organization and what kind of data is flowing into it.
That's what the AI Readiness Assessment gives you: a fast, structured look at where you stand on the five things that turn scattered AI use into a system you can actually trust, starting with the one under this whole question. Data classification. No tooling pitch, just a straight answer to what's safe and what isn't.
"What's safe to paste into ChatGPT?" is a good question. Your team is asking it every day, and answering it themselves. The only question left is whether you've decided first.
Keep reading
Part of a series on AI governance, the structure underneath the tools.
- The Hard Part of AI Was Never the Technology. Why the tool was never the variable.
- The 4 Stages of AI Governance Maturity. Where the data-classification question sits on the larger ladder.