That AI-Generated Image You Made? Anyone Can Take It

A while back, a friend who works in marketing showed me an image.

AI-generated. He typed a prompt, and in a few seconds, out came a poster-quality product shot. He was thrilled: "The photographer quoted me ¥8,000. I did it with one sentence. Free."

I said, sure, the image looks great. But right now, you don't control its fate.

He didn't believe me. You probably don't either. So today, let me clear this up.

One disclaimer first: I talk about marketing, not law. What follows builds general awareness — if you actually get sued, go find a real lawyer.

1. What the Machine Draws Isn't Yours

What is copyright, really?

Simply put: the image belongs to you, and if someone else uses it, you can sue them.

So what about AI-generated images? In March 2026, the U.S. Supreme Court declined to hear Thaler v. Perlmutter, effectively upholding the lower courts' rulings: a work protected by copyright must be created by a human.

That brilliant prompt you wrote doesn't count as authorship. Neither does the "generate" button you pressed.

So that image? Anyone can copy it, modify it, run ads with it. There's nothing you can do about it.

The same goes for text. If you have ChatGPT write a social post and publish it nearly unchanged, that piece is effectively lying in the public domain, free for anyone to use.

I've always said people who make content need a point of view of their own. This is why — your point of view is yours. The machine's sentences are not.

2. What You Substantially Rework Is Yours

So is anything AI has touched a lost cause?

Not quite.

If you make substantial modifications to an AI-generated draft — recomposing the layout, rewriting the copy, injecting your own judgment, so the final piece is genuinely shaped by you — then it can most likely be copyrighted.

But watch out for two traps.

Trap one: how much change counts as "substantial"? There's no bright line. Changing a headline doesn't. Changing it beyond recognition certainly does. And that huge gray area in between? The courts themselves haven't finished arguing about it.

Trap two: if someone challenges you later, you have to be able to prove you made the changes.

How? Keep records. Drafts, edit histories, working files — save them all. Don't go hunting for them after a lawsuit starts; by then they're gone for good.

3. You Can't See What It "Ate"

The third risk is more hidden.

Have you ever wondered what the AI saw before it generated that image?

I don't mean your prompt. I mean its training data — the massive pile of other people's work scraped from the internet.

You use Midjourney to generate an illustration that looks a lot like a certain artist. You don't know. The model doesn't know. But the artist who got imitated knows. And that can become a lawsuit.

Some services are more careful. Getty and Adobe Firefly train only on licensed material, which makes infringement risk far lower. But honestly, what comes out tends to be far more mediocre too. Safe or beautiful — for now, pick one.

Text is a bit better; mainstream chatbots at least cite some of their sources.

And a whole batch of "fair use" lawsuits is queued up in the courts. Whether data scraped from the web needs to be paid for and licensed — until the rulings land, everything is up in the air. I'm watching this closely.

4. What You Feed It, It Will "Eat"

The last risk is one many companies commit daily.

Do you know how that free — or $20-a-month — ChatGPT account of yours actually runs?

By default, every piece of content you type in may end up in its training pool.

You paste in an unreleased proposal for polishing, a client list for analysis, financial figures for a spreadsheet — great, and tomorrow that information may "leak" out through some stranger's question.

What to do? Get the enterprise plan. Enterprise plans build a wall between your team and the training system; your data never enters the training pool.

This is not optional. If your company uses AI to get work done, it's mandatory. If you're using AI on a client's project, there is no room for debate.

And the wall alone isn't enough. Everyone in the company has to understand: sensitive data goes into the enterprise plan only. Double-check what tools your contractors use. No system, however expensive, can stop an employee looking for a shortcut.

A Final Word

Don't get me wrong — I'm not telling you to stay away from AI. Quite the opposite: I'm a heavy AI user, and I genuinely believe every marketer should be using it.

But use it with your eyes open: the rules of this territory haven't finished being drawn. The odds you actually get sued are small, but not zero — especially for big companies, which in a lawyer's eyes are a juicy target.

So: keep your documents, set your policies, buy the right tools. And when you have a moment, ask your insurance broker whether your policy covers AI-related claims. It's a new area — many insurers haven't written it into their terms yet.

The opportunity is the pearl; the risk is the shell. See the shell clearly, and only then is it safe to put the pearl in your mouth.