Copyright, AI outputs and training-data questions

0:00 Copyright and AI

A restaurant client in Dubai sends you a brief: an anime poster, in the style of a famous studio, starring a much-loved cartoon character eating our burger. The client is excited. Your image tool can make it in thirty seconds. Should you? In this lecture you'll learn the three copyright questions every AI user faces, what's settled and what isn't as of 2026, practical safeguards, and how to write contracts that don't promise the impossible.

0:33 Why it matters

Why does this matter? Because AI makes it easy to create things you may not own, and things that may belong to someone else. Get it wrong and you face takedowns, disputes with rights holders, awkward client conversations, and brand assets you can't protect. Get it right and you can move fast with confidence. The law here is still moving. Courts and legislators in many countries are working through it. So this lesson gives you stable principles and habits, and for high-value or contentious work, you get advice from a lawyer in your market.

1:14 The mental model

Here's the mental model. Think of AI output like a photo taken by a security camera. Nobody is the artist. If you then pick the best frames, crop them, arrange them into a collage and paint over parts, you become an author of that collage. Copyright in several major places follows that logic. Purely machine-generated material may not be protected. Your human creative contribution can be. So the three questions are: can I own what AI creates? Is training on copyrighted work lawful? And could my output infringe someone else's rights?

1:54 Q1: Can I own it?

Question one, ownership. In the United States, the Copyright Office's January twenty twenty-five report said prompts alone generally don't make you the author, while human selection, arrangement, modification, and human-written parts can be protected. Courts have upheld the human authorship requirement. The UK has a special computer-generated works provision, but it's under review. Practically, you may not be able to stop others copying purely AI images, so for logos and key brand assets, make sure a human designs or substantially reworks them, and keep records of that work.

2:32 Q2: Is training lawful?

Question two, training. Models learn from huge datasets that often include copyrighted works. In the US, early court rulings in twenty twenty-five reached different answers on fair use depending on the facts, including how the data was obtained, and many cases continue. In the EU, since August twenty twenty-five, providers of general-purpose models must have a copyright policy that respects opt-outs, and publish a summary of training content. The UK hasn't settled the question. For you as a user, most of this risk sits with developers. But choose reputable tools, read their terms, and check whether business plans include any indemnity, and what it actually covers.

3:18 Q3: Could I infringe?

Question three, infringement, and this is where users get into trouble. Your output can infringe if you prompt for a specific character or franchise, if you imitate a named living artist so closely that outputs resemble their works, if you include another brand's logo, if you generate music that closely resembles an existing song, or if you upload someone else's photo or video as an input without permission. The fix is simple: don't name protected things, check outputs for recognizable characters and near-copies, and only feed tools inputs you own or license.

3:58 Example 1: Manchester designer

First example. A freelance designer in Manchester needs a hero image for a coffee brand. Her first prompt mentions a famous illustrator by name. She catches it with her pre-generation checklist, rewrites the prompt to describe qualities instead: loose ink lines, warm muted palette, lots of white space. She uses her own photos of the café as reference inputs, generates, then repaints parts by hand. Before delivery, she runs a reverse image search, finds no near matches, and fills in one row of her provenance record. Ten extra minutes. A much stronger position.

4:38 Example 2 (illustrative): Dubai burger brand

Second example, a realistic business scenario with illustrative details. Back to our Dubai burger brief. The agency explains the risk: a well-known cartoon character is protected by copyright and trademark, and naming a specific studio's style invites near-copies. So they propose something better. Their illustrator creates an original mascot, a cheerful camel chef, and the team uses AI to explore poses and backgrounds, then the illustrator repaints the finals. The contract assigns the illustrator's rights to the client, and the agency's AI clause says it doesn't warrant exclusive ownership of purely AI-generated elements. Illustratively, the client ends up with a mascot it can actually trademark and reuse for years, instead of a poster it might have had to pull in a week.

5:31 Watch me do it: fix a risky prompt

Watch me do it. I'll take a risky prompt and fix it. Here's the original: a poster in the style of a famous animation studio, featuring a well-known mouse character, holding our iced latte. I run the five-line checklist. Named character? Yes, remove it. Named studio style? Yes, rewrite as qualities: hand-drawn look, soft watercolor backgrounds, expressive eyes. Another brand's logo? No. Unlicensed input? I was going to upload a competitor's ad as reference. No, stop. My new prompt: an original, friendly fox barista character, hand-drawn look, soft watercolor background, holding an iced latte in our brand colors. Then I record the tool, plan, inputs and my human edits.

6:18 Contracts

Now contracts. Update your agreements with clients and creators to say whether AI tools may be used and for which deliverables, who owns the outputs, and who checks third-party rights. The key thing for agencies: don't warrant full, exclusive copyright in elements that were generated purely by AI. You may not be able to deliver that. Instead, promise the process: licensed inputs, output checks, records, and assignment of whatever rights you do hold. Clients generally respect honesty, and it keeps you out of promises you can't keep.

6:56 AI music and audio

A special word on music and voice, because this is where creators get caught most often. AI music tools differ a lot in what they let you do commercially. Some free tiers allow personal use only. Some paid plans grant commercial rights to outputs, and some don't. And platforms have their own rules on audio in ads and monetized videos. So before you drop an AI track into a paid campaign, check three things: the tool's current terms for commercial use on your plan, whether the output closely resembles an existing song, and the platform's rules for music in ads. If you're not sure, use the platform's licensed library or a properly licensed track. A copyright claim on a paid ad can switch off your spend in the middle of a launch.

7:53 Common mistakes

Common mistakes. Assuming AI made it, so it's free to use. Promising clients exclusive ownership of AI-only outputs. Uploading a competitor's ads for the model to remix. Using AI music in paid ads without checking the tool's commercial terms. And keeping no records at all, so when a question comes up months later, you can't show what you did. Each of these is avoidable with a two-minute habit.

8:23 Recap + try this now

Recap. Ownership follows human creativity, so document yours. Training lawsuits mostly target developers, but pick reputable tools and read their terms. Your outputs can still infringe if you prompt for protected characters, artists or brands, or upload what you don't own. And contracts should promise a process, not impossible ownership. Try this now: pick one AI-assisted asset you use commercially and fill in one row of the provenance record from the lesson, then add the AI clause to your proposal template. Next: likeness, voice rights and deepfakes.