Copyright & AI Content
Lesson 8.5 — AI and Copyright: What You Need to Know
Copyright and AI is one of the most actively contested legal areas of the decade. Courts in the US, UK, and EU are currently working through cases that will determine fundamental questions: Can AI training on copyrighted data constitute infringement? Who owns AI-generated content? Can you use AI-generated images commercially? The answers are not fully settled, but enough is known to make practical decisions.
This lesson covers the current legal position across major jurisdictions, what you can and cannot safely do commercially, the ongoing training data cases, and practical guidance for real-world use.
Who Owns AI-Generated Content?
The most fundamental question is also the clearest: in most jurisdictions, purely AI-generated content cannot be copyrighted. Copyright requires human authorship.
United States: The US Copyright Office has issued clear guidance (2023): AI-generated content, without sufficient human authorship, is not protectable. The Office has granted copyright in cases where the human creative contribution is significant — for example, a graphic novel that used AI-generated images but had substantial human selection, arrangement, and creative work around them. Pure AI output, without meaningful human creative input, is not protectable.
United Kingdom: UK copyright law requires a human author. The UK Intellectual Property Office has been reviewing how existing laws apply to AI-generated works. Under current law, computer-generated works without a human author may receive a limited term of protection (50 years, under the CDPA 1988), but this is contested and under review.
European Union: The EU AI Act and related copyright guidance make clear that copyright requires human authorship. AI-generated outputs not involving significant human creative contribution are not protectable under EU copyright.
Practical implication: If you generate an image with AI and publish it, a competitor cannot copyright the same image they generate with a similar prompt — but neither can you. The content is effectively in the public domain from a copyright perspective. What you can protect is the business around it, your brand, your human creative curation, and any significantly human-authored elements.
Can You Use AI-Generated Content Commercially?
The question of ownership of AI outputs is separate from the question of whether you are permitted to use them commercially under the tool's terms of service. Read your platform's terms.
| Platform | Commercial use permitted | Conditions |
|---|---|---|
| DALL-E 3 / OpenAI | Yes, for paid users | Subject to OpenAI's usage policies; cannot use to create competing AI systems |
| Midjourney | Yes, for paid subscribers | Free tier: non-commercial only; check your subscription tier |
| Adobe Firefly | Yes | Explicitly designed for commercial use; trained on licensed data |
| Stable Diffusion (base models) | Generally yes | Open weights; check the specific model's licence (CreativeML Open RAIL-M) |
| Google Gemini image generation | Check current terms | Terms evolving; check at time of use |
| Canva AI (Magic Media) | Yes, with Canva Pro | Subject to Canva's content licence |
The safest option for commercial work: Adobe Firefly. It is explicitly trained on licensed and public domain content, and Adobe provides explicit copyright indemnification for commercial use of Firefly outputs. For any work that will appear in client deliverables, advertising, or products, Firefly significantly reduces your legal exposure.
Training Data: The Active Litigation
The most significant legal battles in AI copyright are about training data — the question of whether it was permissible for AI companies to train their models on copyrighted content without permission or payment.
Active cases as of 2025:
Getty Images v. Stability AI: Getty Images (UK and US) sued Stability AI for allegedly copying millions of Getty images to train Stable Diffusion. The case is ongoing. Getty claims the AI reproduces watermarked images and generates images in the style of Getty photographers without licence.
The New York Times v. OpenAI and Microsoft: The NYT sued OpenAI and Microsoft alleging that GPT models were trained on millions of NYT articles without permission. The NYT claims the models can reproduce near-verbatim copies of articles when prompted in certain ways, undermining the Times' subscription business.
Author class actions: Multiple groups of authors (including high-profile names such as John Grisham, Jodi Picoult, and George R.R. Martin) have filed class action suits against OpenAI, Meta, and others, arguing that their copyrighted books were used to train language models without consent or compensation.
The key legal question: Is training AI on copyrighted data "fair use" (US concept) or permitted under other copyright exceptions? AI companies have generally argued yes — that training is transformative and does not reproduce the original. Publishers, authors, and media organisations argue no. Courts will determine the answer, and the outcome will shape how AI is developed going forward.
Key takeaway: The legal picture on training data is genuinely unresolved. What is clear is that AI companies used copyrighted content to train their models. Whether this was permissible, and what compensation (if any) is owed, is being determined in real time.
The "Style" Question
One of the most practically relevant copyright questions for users is: can you prompt AI to generate content "in the style of" a specific artist?
The legal position: Style itself is not copyrightable. You can legally paint in the style of Monet or write prose in the style of Hemingway. The same principle applies to AI-generated work. An AI image "in the style of Impressionism" or "in the style of film noir photography" does not infringe copyright.
The ethical dimension: While legally permissible, generating commercial work "in the style of [living artist]" — effectively using their aesthetic without compensation — is widely considered unethical in the creative community. Many working artists object strongly to this, having spent years developing a distinctive style that now anyone can approximate in seconds.
Practical guidance:
- For personal and educational use: generating work in any style is generally fine
- For commercial work: prefer style descriptions (genre, era, technique) over living individual artist names
- For professional creative work: be aware of and respectful toward the norms of the creative community you're working in
Practical Guidance Summary
| Use case | Recommended approach |
|---|---|
| Personal creative projects | Any major AI tool; no commercial copyright concern |
| Client work or commercial content | Use Adobe Firefly, or verify commercial rights on your specific plan |
| Advertising and marketing materials | Adobe Firefly preferred; consult tool's commercial terms |
| Book or publication illustrations | Adobe Firefly or tools with explicit commercial indemnification |
| Style of living artist for commercial work | Avoid; describe the style without naming the artist |
| Prompting AI with copyrighted text (novels, articles) | Avoid pasting substantial copyrighted text; use summaries or your own words |
What Is Changing
The legal landscape is moving quickly. Potential developments to watch:
- Major training data cases may be settled or decided, establishing precedent or compensation frameworks
- Licensing deals between AI companies and content owners are emerging (OpenAI has signed deals with several publishers)
- Provenance standards (C2PA) may allow AI-generated content to be identified automatically
- New legislation in the EU and UK may create specific AI-related copyright provisions
None of this makes current practical decisions impossible — the guidance above reflects the current best understanding — but it does mean checking the latest before making high-stakes commercial decisions.
Practice Task
Look at the last AI-generated image or text you used for something you shared publicly. Review the tool's current terms of service for commercial use. Was what you did covered? If you're not sure, that uncertainty is exactly what this lesson is designed to address. Know before you publish.