AI copyright disputes generally split into two separate questions that get talked about as if they were one: is it lawful to train a model on copyrighted material, and can the content an AI model produces be copyrighted at all. Both questions are genuinely unsettled in 2026, with courts and copyright offices reaching different, sometimes conflicting, conclusions depending on the specifics. This is general information, not legal advice — copyright outcomes are highly fact-specific and require qualified counsel.
What changed in 2026
- More training-data lawsuits reached substantive rulings. Rather than settling early, several major cases against AI developers produced actual court decisions on fair use, giving the industry real precedent to work from instead of pure speculation.
- Licensing deals accelerated. Publishers, image libraries, and other content owners increasingly signed direct licensing agreements with AI companies rather than pursuing litigation, treating training data as a market to be paid for. See AI training data licensing for how these deals work.
- Copyright offices refined output guidance. Several national copyright offices clarified how much human creative control is needed for AI-assisted works to qualify for protection, moving past the earlier blanket uncertainty.
- Opt-out and consent mechanisms spread. Technical standards letting content creators signal they do not want their work used for AI training gained wider adoption, partly in response to legal pressure and partly to preempt it.
The two questions, separated
| Question |
Current state |
| Is training on copyrighted works fair use? |
Contested; court rulings vary by case facts, jurisdiction, and type of content used |
| Can pure AI-generated output be copyrighted? |
Generally no in the US without meaningful human creative input; hybrid human-AI works may qualify |
| Can AI-assisted output with human editing be copyrighted? |
Often yes, where a human made substantive creative choices, but the human-contributed portions are what is protected |
| Is scraping public web content for training always lawful? |
No — publicly accessible does not mean license-free; site terms and copyright status both matter |
Why fair use arguments are not settled
Fair use analysis in the US weighs several factors — the purpose of the use, the nature of the copyrighted work, how much was used, and the effect on the market for the original. AI training cases have produced different outcomes because courts weigh these factors differently depending on the facts: whether the training data was legally acquired, whether the resulting model can reproduce protected expression, and whether the AI product competes directly with the original market. This is why one favorable ruling for an AI developer does not automatically apply to a different developer with different facts.
Can you copyright what an AI creates?
In the US, copyright protection has generally required a human author, and copyright office guidance has denied protection to output produced with minimal human creative control — prompting alone typically does not qualify. Where a human substantially edits, arranges, or creatively directs AI-generated material, the resulting hybrid work can often claim protection for the human-contributed elements, though the boundary is still being worked out case by case and varies by country.
What this means for businesses using AI-generated content
If your business publishes AI-generated or AI-assisted material, two separate risks apply: the risk that your output infringes on training data the model was built from, and the risk that your own output cannot be protected as your intellectual property if it lacks sufficient human creative input. Practically, this argues for keeping records of human editorial involvement in AI-assisted work, and for AI generated content disclosure practices that keep your usage transparent regardless of how the underlying legal questions resolve.
FAQ
Is it illegal to train an AI model on copyrighted books or images?
Not automatically — it depends on jurisdiction, how the material was acquired, and case-specific fair use or equivalent analysis. Courts have reached different conclusions on different facts, so there is no blanket answer.
Can I copyright a logo or image an AI tool generated for me?
Generally not if you only entered a prompt, under current US guidance. If you substantially edited or creatively directed the output, the human-contributed elements may be protectable — verify current guidance directly.
Do AI companies now pay to license training data?
Increasingly, yes, for major content sources — publishers, stock media libraries, and some other rights holders have signed licensing deals, though scraped open web content remains a separate and more contested category.
Should businesses worry about copyright when publishing AI-assisted content?
It is worth understanding the risk, particularly for close paraphrasing of specific sources or heavily AI-generated visual content resembling existing copyrighted work. When stakes are high, get case-specific legal review rather than relying on general guidance.
Where to go next