Copyright

You've made a logo with AI and you're about to put it on everything. Do you own it? Nowhere has fully settled this. But the systems all rest on the same idea, and that idea points somewhere useful. I'll use Australia as the worked example, and tell you where the United States and the United Kingdom sit differently. And I'm not a lawyer — if this matters commercially, get one who knows your part of the world.

Almost every copyright system has a person at the centre of it. Take Australia. No AI legislation, no decided case. But two judgments from around twenty ten — one about a television guide, one about a phone directory — settled that a piece of work needs a human author who made an original intellectual contribution. Neither had generative AI anywhere near it. That requirement isn't unique to Australia; it runs through most systems in some form. So it's tempting to ask who owns AI output. The question underneath is whether copyright exists in it at all — because if it doesn't, there's nothing to own, and you couldn't stop a competitor using the same image or a stranger printing it on t-shirts.

Three big English-speaking systems, three starting points. Australia has nothing on the books and nothing decided. The United States has come closest to saying it outright: its Supreme Court declined to reopen the question in March twenty twenty-six, leaving standing the position that purely AI-generated work has no owner. The United Kingdom is the odd one out — a decades-old provision there does protect computer-generated work with no human author, and a government report in the same month recommended getting rid of it, while keeping protection where a person made real creative choices. All moving the same way.

So it's a spectrum, not a switch. At the weak end: one prompt, take the first thing it gives you, publish it unchanged. Very little of you in that. At the strong end: AI as one tool inside a process you're driving — you direct it, reject twenty versions, rework the one you keep, combine it with your own material. That's much closer to how copyright has always understood authorship, in all three systems.

Now, the terms of service. Most of these tools say you own what you create with them, and people read that as settled. It's a contract between you and that company — a promise they won't claim it and won't come after you. But a contract binds the two parties who signed it, and copyright binds everyone. No set of terms can hand you a right the law doesn't recognise. The company can't give away something it never had. There is one promise worth more than the others. Adobe, Microsoft and Google will each defend you — cover the legal costs if someone claims the image you generated infringes theirs. Read the plan you're on, because that cover is usually business tiers only. Adobe's version goes furthest: their Firefly model is trained on licensed and public-domain material, so there's less to be claimed in the first place. None of that gives you copyright. It moves the risk off your desk, which is a different and useful thing.

So a working rule, scaled to what's at stake. A social post nobody will copy — use whatever works, don't think about it. Something you're charging a client for — keep some record of the direction and the choices you made. A logo, a brand asset, anything you'd want to stop someone else using: start with AI if it helps, but don't finish with it. What you add at the end is what you own.

One thing to do this week. Look at whatever you've made with AI that's currently doing a commercial job for you — a logo, a header image, something on your website. Ask yourself how much of you is in it. If the answer is one prompt and no changes, that's worth knowing now rather than the day someone else starts using it too.