Somewhere in the last two years, “just AI it” became a common creative shortcut. Need a hero image for a landing page? Prompt it. Need a background extended, a product shot cleaned up, or an object removed from a photo? Use AI. Need concept art for a campaign nobody’s approved the budget for yet? Prompt it, obviously.
It’s fast, and depending on your org, your tools, and how much you’re using them, sometimes cheaper, too. It’s also occasionally unsettling in how good it’s gotten. What almost nobody asks is a much simpler question: once you have the image, do you legally own it?
We got curious about this ourselves and spent some time in the court filings, the Copyright Office rulings, and the ongoing lawsuits, so your team doesn’t have to. What we found was more complicated than we expected, and the landscape is still moving quickly enough that some of this could shift again before the year is out. Here’s what’s real, what’s still unsettled, and what to do about it.
A quick disclaimer: We’re not attorneys, and nothing below is legal advice. We became curious about how AI is reshaping creativity and our industry, so we did the digging. Talk to your own attorney before making any intellectual property decisions about your AI-generated content.
The Short Answer: Probably Not, But It Depends
Whether or not your AI output is copyrightable depends on how the image was made. Here are a few different scenarios.
If the image was generated entirely from a prompt
It’s probably not copyrightable, at least not the way you’d hope.
In the US, copyright has always required a human author. That’s not a new AI-era rule. It goes back to a 19th-century photography case that established a camera operator’s creative choices could earn copyright protection, machine or no machine. Courts have consistently held that a full machine-generated image, one where a human didn’t meaningfully direct or shape the results, doesn’t clear that bar.
Where it gets murkier, however, is everything in between. The Copyright Office has rejected claims from artists who argued that skillful, extensive prompting was itself the creative act. At this time, there isn’t a clear answer for where the line sits between “AI did this” and “a human directed this using AI.”
So, if your “proprietary” hero image came entirely out of a prompt box, you may not have exclusive rights to it. Someone else could generate something nearly identical, and you’d have no real legal ground to stand on.
If a human heavily directed or edited the AI’s output
It’s still unclear if it’s copyrightable. This is the unsettled part that’s actively being tested in court right now.
There’s a more forgiving nuance in the Copyright Office’s own guidance. AI-generated portions of a work only lose protection to the extent those portions would’ve been copyrightable on their own. So, using AI for something that wouldn’t be independently copyrightable anyway, such as extending a background or increasing resolution, doesn’t sink the copyrightability of the piece as a whole. You still have to disclose to the Copyright Office that AI was used, but that’s a separate requirement from whether the finished result holds up as copyrightable.
Beyond that narrow case, though, nobody has a clean answer yet for how much human direction is enough. That question is sitting in front of a federal court in Colorado right now, in a case worth watching in Allen v. Perlmutter (more on that below).
If you used AI to edit an existing photo you licensed
It’s most likely not copyrightable, and it’s important to check the terms in your tool’s licensing agreement.
Generative Fill, background extension, object removal, and img2img tools alter a photo that already exists, and that photo might comes with a license attached. A license that allows editing for web use doesn’t automatically include AI editing and redistribution. Some newer licenses say so directly. Most older ones don’t mention it, which means you cannot assume it does.
Purpose-built tools like Adobe Firefly or Getty’s own generative AI reduce your risk of a claim from the source content owner, especially when that same company is giving you both the image and the tool used to alter it, and some come with indemnification on top of that. But that reduces litigation risk, it doesn’t answer the copyrightability question, and indemnification usually covers your use of the tool, not the output itself. Feeding a licensed photo into some other, unrelated AI tool carries a different risk, since it doesn’t come with that same protection.
Before publishing anything AI-edited, ask yourself two things: was this allowed, and did I do enough to make it mine?
The Tool Itself Might be a Problem
Even if you land on solid ground for authorship, there’s a separate issue lurking underneath. Did the AI tool infringe someone else’s copyright to get good enough to make your image in the first place?
One theory says the AI company itself infringed by training its model on copyrighted images without permission. A separate theory says the company is liable because it built and promoted a tool that lets users infringe. The distinction matters because a court could decide that training an AI model counts as fair use, meaning the company didn’t infringe by building the tool, without that protection extending to what a specific output looks like. And even if the AI company isn’t liable for what users generate, that doesn’t mean users are automatically in the clear.
It’s the same logic courts have applied to VCRs for decades. The manufacturer isn’t liable just because someone used the machine to copy a movie, but the person who copied and distributed that movie still is. If your team generates an image that closely resembles a specific piece of existing copyrighted art, there may not be much to defend, since the exposure could sit with whoever generated and used the output.
Furthermore, “fair use” isn’t finding that no infringement occurred. It’s a legal defense that excuses infringement that did occur, under specific conditions. A ruling that AI training is fair use is still saying the training technically infringed copyright, but that infringement is legally permitted. This is important if you’re trying to gauge how protected your own use of these tools is.
Cases to Watch
Here’s the litigation behind everything above, grouped by which question each one is testing.
Testing what counts as enough human authorship:
- Thaler v. Perlmutter — The Supreme Court declined to hear an appeal in March 2026 from a computer scientist who listed his AI system as the author of an image it created entirely on its own. The lower courts had already said no, and the Court's silence left that ruling as the controlling case on the subject. However, because Thaler claimed the AI was the author rather than himself, the case never tested what happens when a human lists themselves as author and describes AI as a tool they used along the way.
- Allen v. Perlmutter — That untested scenario is what's now in front of a Colorado federal court. Artist Jason Allen is suing the Copyright Office over its refusal to register an image he made with over 600 Midjourney prompts and Photoshop edits, an image that won first place at the Colorado State Fair. Unlike Thaler, Allen listed himself as the author. The case is fully briefed, with a ruling expected sometime this year, making it the one to watch for a more direct answer.
Testing whether the tools themselves infringed:
- Disney and Universal v. Midjourney — Disney, Universal, and several other studios are suing Midjourney directly, alleging the platform was trained on its copyrighted characters and keeps generating derivative images of them on request. The case is now deep in a contentious discovery fight. Midjourney is even pushing to force the studios to disclose their own use of AI. Nothing is decided yet, but the stakes and scale of the litigation show how seriously the courts are taking it.
- Andersen v. Stability AI — This is the original class action from a group of working artists, filed back in 2023 over a dataset of five billion scraped images. It’s headed to trial in September 2026, and it’s the case most likely to produce a real answer on whether training an image model on copyrighted art without a license counts as fair use or infringement.
None of these cases have delivered a final, universal ruling yet. What they have delivered is a very clear signal that the tools generating your visuals are themselves tangled in active litigation, and the outcome could change what you’re allowed to do with what they produce.
So, What Should You Do About This?
Again, we are not attorneys, and none of this is legal advice. These are simply a few practical takeaways from what we found:
Treat “AI-touched” and “brand asset” as two different categories.
A raw output from a prompt, or a lightly-altered stock photo, is a draft, not a deliverable. The more a human edits, composites, art-directs, and makes deliberate creative choices on top of the AI’s contribution, the stronger the case. But again, this is very nuanced.
Read the fine print on your tools.
Using the source’s own AI tool reduces litigation risk, but doesn’t guarantee copyrightability. A licensed stock photo run through an unrelated AI editing tool carries a different risk. Know what your license and your tool’s terms cover before you build a campaign around it.
Assume nothing is exclusive.
If a visual isn’t copyrightable, you don’t have exclusive rights to it. That means there’s nothing stopping someone else from using that same image or generating something close to it. Build your campaigns and brand identity on assets you can defend.
If you’re an agency, check your contracts on both ends.
You can’t transfer ownership rights you don’t have, so if an AI-generated image isn’t copyrightable in the first place, there’s nothing to assign to your client no matter what your contract says. Client agreements are also increasingly adding their own limits or outright bans on AI use in deliverables. It’s important to ensure everyone on the account knows which projects have restrictions.
Label your synthetic content if you’re marketing into Europe or New York.
The EU’s AI Act requires disclosure when content is AI-generated, even if your company isn’t EU-based, as long as you’re reaching audiences there. This means a label on the content itself, wherever it’s published. New York has its own narrower version, and as of June 2026, ads reaching New York consumers with an AI-generated “synthetic performer” require conspicuous disclosure, with real penalties. However, this is all very nuanced and evolving rapidly.
Frequently Asked Questions
Is AI-generated art copyrighted?
No, not automatically. US copyright law requires a human author, and courts and the Copyright Office have consistently ruled that fully AI-generated images don’t qualify, regardless of how much prompting went into them.
Who owns an image created by AI?
Nobody holds exclusive copyright over a purely AI-generated image, since it doesn’t meet the human authorship requirement. If a human makes substantial creative edits on top of the AI output, that added work may be copyrightable, but the underlying AI-generated portion generally isn't.
Can I use AI-generated images for my business?
Yes, in most cases. Some laws, like the EU’s AI Act, may require disclosure depending on your audience. Exclusivity is a separate issue. If the content isn’t copyrightable, a competitor can use something similar, which is more of a business risk than a legal one. You can still face an infringement claim on the image itself, even if the tool was never found to have infringed anything during training. If a tool offers indemnification, check whether it covers your use of the tool and/or the output it generates.
Do I need to disclose that an image is AI-generated?
It depends on where your audience is. The EU’s AI Act requires disclosure, as of August 2, 2026, for AI-generated content reaching audiences in the EU, regardless of where your company is based. NewYork has its own narrower rule, requiring disclosure for AI-generated “synthetic performers” in ads reaching New York consumers as of June 2026. The US doesn’t currently have a comparable federal requirement, though that could change.
Is it legal to edit a licensed photo with AI tools like Generative Fill?
It depends on your license. AI-editing a licensed image can create a derivative work that falls outside what the original license permits. Some newer stock licenses address AI editing directly. Older ones often don’t mention it at all, which means the answer is worth checking rather than assuming.
Is there a court case testing whether a human can copyrightAI-assisted art?
Yes. Allen v. Perlmutter is currently testing this directly. Unlike an earlier case where the AI itself was listed as the author, artist Jason Allen listed himself as the author of an image he made using extensive Midjourney prompting and Photoshop editing. The CopyrightOffice refused to register it anyway. The case is fully briefed in Colorado federal court, with a ruling expected sometime in 2026.
Are there lawsuits over AI image generators?
Yes, several are ongoing. Getty Images sued Stability AI over training data (the UK court largely ruled against Getty’s copyright claim in late 2025). Disney and Universal are suing Midjourney over character reproduction. Andersen v. Stability AI, a class action from working artists, is headed to trial in September 2026. None have produced a final, universal ruling yet.
