Copyright and ownership of AI images

0:00 Copyright and ownership

A client calls you, excited. They love the AI mascot you generated last week, and they want to put it on everything. Packaging, merchandise, a trademark filing. Then they ask a simple question. We own this, right? What do you say? If you say yes, of course, you might be wrong. And if a competitor copies it, your client may find out the hard way. In this lecture, you'll learn the three separate questions behind AI copyright, what the US Copyright Office and courts have said, how the UK and other markets differ, and a practical workflow that makes key assets far more protectable. This isn't legal advice, but it will make you a much safer partner for your clients.

0:52 Why it matters

Why does this matter? Because ownership is value. A brand invests money into a logo or mascot precisely because it can stop others from copying it. If a key asset can't be protected, that investment leaks. And the rules around AI are still developing, differ between countries, and are often misunderstood. Many people believe that if a tool's terms say you own the output, then it's protected. That's a different question. Think of it like buying a house. The seller can hand you the keys, but that doesn't mean the house has planning permission. Tool terms are the keys. Copyright law decides whether there's anything to protect.

1:39 Three separate questions

Here's the framework. There are three separate questions, and people mix them up constantly. Question one: is the output itself protected by copyright, and who owns it? Question two: might the output infringe someone else's rights, like a character, an artwork or a trademark? Question three: what do the tool's terms say about using the output, and about your inputs? You can get a yes on one and a no on another. For example, the terms may let you use an image commercially, it may not infringe anything, and yet it may have little or no copyright protection. Keep these three questions separate, and most confusion disappears.

2:26 United States

Now, the United States, where the guidance is clearest. In January twenty twenty-five, the US Copyright Office published part two of its report on AI, focused on copyrightability. Its conclusion: AI outputs can be protected only where a human author has determined sufficient expressive elements. That can include a human's own expressive work that's perceptible in the output, or creative selection, arrangement or modification of AI material. But prompts alone, based on the technology at the time, didn't give enough control. The courts agree on human authorship. In the Thaler case, an image listing an AI system as its author was refused, and in March twenty twenty-six, the Supreme Court declined to hear the appeal. So the human authorship requirement stands.

3:19 Other markets

What about elsewhere? In the United Kingdom, there's an unusual provision for computer-generated works with no human author. It gives authorship to the person who made the arrangements for the work to be created, with a shorter term. How that applies to modern generative AI is debated, and the government's twenty twenty-six report on copyright and AI discusses the criticism. So treat it as unsettled. In the European Union, protection generally requires the author's own intellectual creation, which points to human creative choices. And Pakistan, the UAE, Saudi Arabia and other markets have their own copyright laws with developing positions on AI. The practical takeaway is the same everywhere. More human creative contribution means more protectable elements.

4:10 Worked example 1: a book cover

Worked example one, simple. A book cover. An author generates a background landscape with AI, then a designer sets the title in a carefully chosen typeface, arranges the composition, paints over the sky by hand, and adds a hand-drawn illustration of the main character. What's protectable? The raw generated landscape, by itself, likely isn't under the US approach. But the designer's typography layout, the hand-painted changes, the illustration, and the overall selection and arrangement are human contributions. If the author ever registers the cover in the US, they'd disclose the AI-generated background and claim the human-authored parts. The key idea: document what the humans did.

4:56 Worked example 2: a startup mascot

Worked example two, the mascot, a business scenario with illustrative details. A snack startup in Riyadh wants a mascot for packaging and a trademark. Before: they pick one raw generation and start printing. That's risky. The image may lack copyright protection, and nobody checked whether it resembles an existing character. After: the agency uses AI for about forty concept explorations. The founder picks a direction. Then a human illustrator redraws it by hand as clean vectors, refines the expressions, and builds a character sheet. The agency runs reverse image and trademark searches, keeps a human authorship record, and assigns rights to the client in writing. Same starting idea, a much stronger asset.

5:45 Watch me do it: an authorship record

Watch me do it. I'll fill in a human authorship record for that mascot. Section one, concept and brief: I attach the founder's notes and my first pencil sketches, with dates. Section two, AI use: I paste the prompt log entries and list the reference images. Section three, selection: forty options explored, three shortlisted, one chosen, with the reason. Section four, arrangement: the pose and composition decisions. Section five, modification: the illustrator's full vector redraw, with the layered master file showing the work. Section six, other human elements: the wordmark typography. Section seven, clearance: reverse image search and a trademark search, with dates. It takes fifteen minutes and could matter for years.

6:34 Infringement and terms

Now, infringement and terms. Even if you can use an output, it can still infringe. Prompts naming characters, films or living artists can produce close copies. Outputs can include pseudo-logos that resemble real brands. So avoid those prompts, run a similarity check on important outputs, and keep your prompt log as evidence of independent creation. There are also ongoing lawsuits about training on copyrighted works, which mostly affect model providers. And then the terms. They may grant you rights only on paid plans, give the provider a license to your inputs, restrict certain uses, and require you to own what you upload. Save a copy of the terms in force when you create client work.

7:24 Common mistakes

Common mistakes. Assuming that because the tool says you own it, it's protected. Assuming AI outputs can't infringe. Ignoring plan-specific terms. Using purely generated images as core brand assets, like logos and mascots, without human authorship. Forgetting to disclose AI material when registering a work. And leaving ownership out of the contract. Your client contracts should say that AI tools may be used, assign whatever rights you hold, note that purely AI-generated elements may have limited protection, and set out who is responsible for what. Get a lawyer to review your template once. It's cheap compared to a dispute.

8:07 Recap and try this now

Let's recap. Keep three questions separate: protection, infringement and terms. In the US, AI outputs are protected only where humans contribute sufficient expressive elements. Prompts alone weren't enough, and the Supreme Court left the human authorship requirement in place. The UK position is unsettled, and other markets are developing. So for key assets, use AI to explore, and let humans create and refine the final work. Document it, run clearance searches, and put AI terms in your contracts. Try this now. Open the terms of one image tool you use and summarize, in five bullets, what they say about output use, input rights, training, restrictions and indemnity. Then start a human authorship record for your most important asset.