Generative AI models learn from enormous collections of text, images, music and code — much of it protected by copyright. Whether that training is lawful without permission is one of the most consequential legal questions in technology. Dozens of lawsuits are working through courts, and the answers may differ between countries.

The core questions

  1. Training: is copying works to train a model an infringement, or is it permitted — for example as fair use in the US or under text-and-data-mining exceptions in the EU?
  2. Outputs: when a model produces text or images closely resembling protected works, who is liable?
  3. Acquisition: does it matter how training data was obtained, for example from pirated sources?

The United States: fair use

US law allows some uses of copyrighted works without permission under fair use, judged case by case on four factors: the purpose and character of the use (including whether it is “transformative”), the nature of the work, the amount used, and the effect on the market for the original.

AI developers argue that training is transformative: models learn statistical patterns rather than storing copies for people to read. Rights holders argue that AI outputs compete with their works and that training copies entire works commercially.

Key cases include:

  • The New York Times v. Microsoft and OpenAI (filed December 2023), alleging that millions of articles were used without permission and that models can reproduce them.
  • Authors’ lawsuits against several AI companies over books. In 2025, a US federal judge in a case against Anthropic found that training on lawfully acquired books could be fair use but treated the use of pirated copies differently; the case was followed by a large settlement with authors.
  • Visual artists and image libraries, including Getty Images, suing image-generator companies.
  • Music publishers and record labels suing AI companies over lyrics and songs.

Different judges have reached different conclusions on parts of these questions, and appeals are likely, so the law remains unsettled.

The European Union: text and data mining

The EU’s 2019 Copyright Directive includes exceptions for text and data mining. For commercial purposes, mining is allowed unless the rights holder has opted out in a machine-readable way. The AI Act requires providers of general-purpose AI models to have a policy to respect these opt-outs and to publish a summary of training content.

Other jurisdictions

Japan’s copyright law includes a broad exception for data analysis. The United Kingdom has consulted on reforms. Approaches differ, which complicates global training practices.

Licensing as an alternative

Many AI companies now sign licensing deals with publishers, image libraries and platforms — see our explainer on content licensing deals. Deals reduce legal risk and provide fresh content, but terms are rarely public, and smaller creators often lack bargaining power.

What creators can do

  • Use opt-out mechanisms where available, such as robots.txt rules for AI crawlers and machine-readable reservations.
  • Register works where registration strengthens legal remedies.
  • Follow collective initiatives by industry associations.

Why outcomes differ by country

Copyright exceptions differ: US fair use is decided case by case, while the EU has text-and-data-mining rules with opt-outs.

What to watch next

  • Appeals court rulings on fair use for AI training in the US.
  • Implementation of training-data transparency summaries under the EU AI Act.
  • Standards for machine-readable opt-outs.
  • Collective licensing schemes for creators.

Key takeaways

  • The key questions concern training, outputs and how data was obtained.
  • In the US, fair use is decided case by case, and courts have not yet settled the issue.
  • The EU allows commercial text and data mining unless rights holders opt out.
  • High-profile lawsuits involve news publishers, authors, artists and music companies.
  • Licensing deals are growing as a way to reduce legal risk.

Frequently asked questions

It depends on the country, the circumstances and ongoing court decisions. In the US it turns on fair use; in the EU on text-and-data-mining exceptions and opt-outs.

Can I stop AI companies from training on my website?

You can block known AI crawlers in robots.txt and use machine-readable rights reservations. Not every crawler respects them.

If an AI output copies a work, who is liable?

Potentially the user who publishes it and possibly the provider, depending on the facts and jurisdiction. Courts are still developing the rules.