Copyright and AI is one of the most contested legal questions of the current moment. The short answer: content generated entirely by an AI system, without meaningful human creative input, is generally not protected by copyright in the United States and most of Europe. But the moment a human makes substantive creative choices — selecting, arranging, or editing AI output in an expressive way — that human contribution may be protected. The AI’s contribution still is not.

Why pure AI output usually has no copyright

Copyright law in the US and EU requires a human author. The US Copyright Office has stated clearly that it will not register works produced by machines operating without human creative control. A 2023–2024 federal court case — Thaler v. Perlmutter — confirmed this: the DC Circuit ruled that AI cannot be listed as an author, and that a work created autonomously by an AI system is not eligible for copyright protection.

The same principle applies in the EU: the EU Copyright Directive requires that a work reflect the author’s own intellectual creation. A machine has no personality, no intention, no creative expression — so it cannot author a work under current law.

This means a business that generates text, images, or music entirely through an AI prompt and uses it commercially has limited legal protection against copying. Anyone who copies that output may not be infringing copyright, because there was no copyright to infringe.

When human creativity changes the picture

The analysis changes when a human makes meaningful creative choices on top of AI output. The US Copyright Office has registered works where a human:

  • Arranged AI-generated images in a creative sequence (as in a graphic novel)
  • Substantially edited or transformed AI text into an original expression
  • Selected specific AI outputs from many possibilities in a creatively purposeful way

In those cases, the human-authored selection and arrangement may be protected — but only that part. The underlying AI-generated elements remain unprotected.

The takeaway for creators: document your creative decisions. The more you can show that you made expressive choices — not just typed a prompt and used the output — the stronger your claim to protection.

The training data debate

A separate and unresolved question is whether AI companies infringed copyright by training their models on copyrighted text, images, and code scraped from the internet.

In December 2023, The New York Times sued OpenAI and Microsoft, alleging that training large language models on its journalism without permission constitutes copyright infringement. Getty Images has sued Stability AI over the use of its photo library to train image generation models. Thousands of authors have filed class-action suits against AI companies.

The AI companies argue that training on publicly available data qualifies as fair use — a US doctrine that permits certain uses of copyrighted material without permission, particularly for transformative purposes. Courts have not yet issued definitive rulings on the merits. The outcome of these cases will shape the legal foundation of the entire generative AI industry.

In the EU, the situation is somewhat different: the EU Copyright Directive includes a text and data mining exception that allows mining for research, with a commercial exception that rights holders can opt out of. Several European publishers have done exactly that.

What businesses should know

For most day-to-day uses — drafting emails, generating marketing copy, summarizing documents — the copyright question is manageable:

  • Content you generate likely has weak or no copyright protection unless a human has made substantial creative contributions. This may matter less than you think for internal use, but consider it for content you plan to assert ownership over publicly.
  • Vendor terms vary. OpenAI, Anthropic, Google, and others assign output rights to users in their terms of service, but those contractual rights are not the same as copyright. Read what your vendor actually grants.
  • Training-data liability is still being litigated. Businesses building AI products on top of third-party models inherit some exposure to how those models were trained.
  • Disclosure requirements are emerging: some jurisdictions now require labeling AI-generated content in certain contexts (advertising, news, electoral communications).

In the news

The intellectual property stakes of AI became concrete in June 2026, when Anthropic revealed that a large-scale campaign had extracted nearly 29 million conversations from Claude without authorization — with the explicit goal of replicating Claude’s capabilities in a competing model. The incident illustrates why AI companies treat their models’ outputs as commercially sensitive, regardless of the unsettled state of copyright law. → Read the full story

FAQ

Can I copyright an image made with Midjourney or DALL-E?
Generally, not the AI-generated portions. The US Copyright Office has declined to register works produced entirely by text-to-image AI. If you substantially edited the output or made creative selections across many generated images, the human-authored parts of that process may qualify.

Do AI companies own the content I generate with their tools?
Not typically — the major providers assign output rights to the user in their terms of service. But those are contractual rights, not copyright, and they do not give you protection against third parties who copy the output.

Is it legal to train an AI on copyrighted material?
This is actively litigated. US courts are weighing whether training constitutes fair use. No binding appellate ruling has settled the question as of mid-2026. In the EU, rights holders can opt out of commercial text-and-data-mining, and some have.

What happens if a country decides AI can be an author?
A handful of countries — notably China and the UK, under certain conditions — have shown more openness to protecting AI-generated works. If those regimes persist and are tested in court, the result would be a patchwork: AI output protected in some jurisdictions, not others. For global businesses, that creates complexity.

Sources: Artificial intelligence and copyright — Wikipedia; US Copyright Office guidance on AI and copyright (2023–2024); Thaler v. Perlmutter, D.C. Circuit (2024).