Copyright: who owns the text, image and code the AI made

In the last lesson — on privacy — you learned to think about what's safe to give an AI. This lesson is the other side: what happens to what the AI gives you. You generated a great picture, text or piece of code — but is it yours? The answer surprises even experienced users.

Who's the author — the person or the machine?

Copyright is a creator's exclusive right to decide what happens to their work: to allow copying, selling and modification, and to forbid it to others. The key word here is a human creator. It varies by country, but broadly, offices and courts work from the position that copyright protects human creativity, not a machine's output.

According to media reports, in 2022–2023 the copyright office in the US refused in a number of decisions to register images created by a neural network — on the grounds that the work has no human author. In one well-known case, an AI-generated painting reportedly won a prize at an art competition — and immediately set off an argument about whether it can count as an authored work at all. Check the exact details of these cases in the original sources, but the direction is clear: in many countries, pure AI output may get no protection at all.

An important point about countries. There's no single global rule: the position of the US Copyright Office may differ from the approach in the European Union or elsewhere. Always check the rules of your own country — and for a serious commercial project, ask a lawyer.

So the picture belongs to nobody?

Here's the unexpected conclusion. If a work isn't protected by copyright, it sits closer to the public domain — the stuff anyone is free to use. Which means the picture an AI drew from your prompt could, in some countries, theoretically be picked up and used by anyone — your competitor included — and stopping them won't be as easy as stopping someone copying a photo you took yourself.

A service's terms are not the law

This is where beginners get muddled. Copyright (your country's law) is one thing; the terms of use — the contract you accept with a tick-box when you sign up — are another. Many AI services explicitly allow you to use what you generate, commercial use included. But that's permission from a company — a licence (the owner's permission to use something), not a guarantee that the law recognises you as the author.

And the terms differ from service to service: some hand the rights to the output to the user, some keep restrictions, some have different rules on free and paid plans. The rule is simple: read the terms of the service you actually use, especially the section on rights to the output.

The risk of resembling someone else's work

Remember the lesson on how a network learns: the model grew on a gigantic pile of text and pictures from the internet — its training data. Some of that data is other people's protected work. Usually AI creates something new (generation, not copying — remember the LLM lesson), but sometimes the result can come out too similar to a specific existing work, to an author's recognisable style, or to a trademark (a brand's name or logo).

Using a result like that risks copyright infringement (using someone's work without permission), and you're the one who answers for it, not the AI. Reportedly, the very fact of training on other people's data is already the subject of lawsuits: stock-photo libraries and authors are bringing claims against model developers. How they end, time will tell — what matters to you is the practical takeaway below.

What about the rights to my prompt?

A prompt (your request to the AI from earlier lessons) is text you wrote. On its own, your prompt is your creative work, like any sentence you write. The output is trickier: the more of your human contribution is in it — edits, choosing the good options, arranging, refining — the better your chance that some countries will recognise your authorship of the finished work. Press one button and take the answer as it comes, and your contribution is minimal.

Refine it at length, assemble it from pieces, rewrite it in your own voice, and your contribution carries weight. It's not an iron-clad guarantee, but it's the principle many countries are leaning toward.

Practical takeaways for a beginner

Here it all is as a short summary — what's safe, what's risky, and how to cover yourself:

• Safe: using generated material for drafts, ideas, inspiration and internal work; refining the result with your own hands; reading a service's terms before you use anything commercially.
• Risky: putting pure AI output at the heart of a logo or a brand and counting on exclusive rights; publishing a result that resembles a recognisable work, style or brand; passing generated work off as human-made where that matters — to a client, in a competition, under a contract.
• How to cover yourself: put your own work in — edit and combine rather than taking it "as is"; check the result for accidental resemblance (drop the image into Google's image search, search for a distinctive phrase); keep your service's terms to hand; for a serious commercial project, an hour with a lawyer is cheaper than the dispute afterwards.

The moral, in the spirit of the whole course: AI is a powerful tool, but responsibility for the result stays with the person. Same as with hallucinations and privacy: it isn't the machine that answers, it's whoever didn't think about where the result came from and where it's going next.

Do this now

An exercise with no tools — just your head and a sheet of paper (or your notes app). Imagine you've generated three things: (1) an image for the logo of your future business, (2) the text of a social media post, (3) a chunk of code for a work project. For each one, answer three questions in your own words: who does it belong to, in principle? what would be risky to do with it? how would I cover myself before using it commercially? Write the answers down. That's the habit of legal caution — worth more than any generated masterpiece.