AI and Copyright: The Key Principles You Should Know

Copyright protects original works created by humans, and AI challenges that principle on two fronts: who owns content produced with AI, and whether AI models may learn from copyrighted works. Here are the key principles, based on official texts, to help you find your way before you publish, sell or protect a creation.

This article is general information. It is not legal advice and does not replace a consultation with a qualified intellectual property attorney, who can assess your specific situation.

Why the question comes up

Generative AI produces text, images, music and video from a prompt. To do so, it was trained on vast amounts of existing content, its training data, some of which is protected by copyright.

That raises three separate questions:

1. Output: is AI-assisted content protected, and who owns it?
2. Input: is it lawful to train an AI on copyrighted works?
3. In between: what happens when the result looks too much like an existing work or imitates a real person?

The answers vary by country. Here we rely mainly on the work of the US Copyright Office, which has published a detailed analysis, along with UK and European Union law.

Question one: is a creation made with AI protected?

The principle: a human author

In the United States, the Copyright Office published Part 2 of its report on copyright and artificial intelligence, on the copyrightability of AI outputs, on January 29, 2025 (official report page, accessed October 22, 2026). It also issued registration guidance for works containing AI-generated material on March 16, 2023.

The United Kingdom takes a distinctive approach. Section 9(3) of the Copyright, Designs and Patents Act 1988 provides that for a computer-generated work, “the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken” (legislation.gov.uk, accessed October 22, 2026). How that rule applies to modern generative AI is still debated.

What this means in practice

The table below applies the Copyright Office’s principles to common situations. Courts in other countries may reason differently, and the outcome always depends on the facts.

Situation	Copyright protection
Image from a single prompt, no edits	Unlikely
Text you wrote, only proofread by AI	Yes, for your text
Generated illustration heavily reworked by hand	Yes, for your creative edits
Collection of generated images you selected and arranged	Possible, for the selection and arrangement
Your drawing transformed by AI, still recognizable	Yes, for the elements of your drawing

The practical consequence: if content is not protected, you can often use it, but you will struggle to stop other people from copying it. So keep records of your own work, such as drafts, versions and edits. That record is what supports your rights.

Check yourself You type a detailed prompt and the AI produces a poster that you publish as is. Are you the author of the poster under US copyright law?

Probably not, according to the US Copyright Office: a prompt, even a detailed one, does not give enough control over the result. If you rework the poster, add your own text or your own photo, those creative contributions can be protected.

Question two: training on copyrighted works

In the United States: a fair use debate

In the US, the question largely turns on “fair use”, an exception assessed case by case. The Copyright Office released a pre-publication version of Part 3 of its report, on generative AI training, on May 9, 2025. Several lawsuits also pit rights holders against AI developers.

In the European Union: an exception with an opt-out

The EU copyright directive 2019/790 provides, in Article 4, an exception for “text and data mining”: lawfully accessible content may be analyzed automatically unless the rights holder has reserved their rights in an appropriate way, for example by machine-readable means for online content.

The EU AI Act builds on that rule. Its Article 53 requires providers of general-purpose AI models to put in place a policy to comply with EU copyright law, in particular to identify and respect those reservations, and to publish a sufficiently detailed summary of the content used for training, following a template from the AI Office (Article 53 on the European Commission’s AI Act Service Desk, accessed October 22, 2026). According to the European Commission, these obligations have applied since August 2, 2025.

On July 10, 2025, the Commission published a voluntary code of practice with a chapter on copyright. Signatories commit, among other things, not to circumvent subscription models or paywalls, to use web crawlers that follow robots.txt instructions, to mitigate the risk of infringing outputs and to offer rights holders a way to file complaints.

Question three: similarity, style and real people

Generated content can reproduce, in whole or in part, an existing work: a famous character, a passage of text, a melody. If it copies protected elements, publishing it may infringe copyright, even if you did not intend it. The person who publishes is responsible for checking.

Style, on the other hand, is generally not protected as such: copyright protects expression, not ideas or a manner of working. In Part 1 of its report, on digital replicas, the Copyright Office did not recommend extending new protections to artistic style. Imitating “in the style of” is therefore not automatically unlawful, but the closer the result gets to specific works, the higher the risk.

Finally, copyright is not the only law at stake. A person’s likeness and voice are protected by other rules, such as state rights of publicity in the US, and a deepfake published without consent can lead to legal action. Our guide to AI music and AI voices covers the question of consent.

Transparency: labeling generated content

In the EU, Article 50 of the AI Act requires synthetic audio, images, video and text to be marked in a machine-readable way, often with an invisible watermark, and requires deepfakes to be disclosed. According to the European Commission, these transparency rules take effect in August 2026. To learn how to spot such content yourself, read how to recognize an AI-generated image, and for an overview of the regulation, see our AI Act summary.

Good habits before you publish

• Read the service’s terms: they set out who may use the output and for what.
• Avoid naming a specific work or living artist in your prompt to get a copy.
• Check for similarity with well-known works, especially characters, logos and melodies.
• Document your work: prompts, versions, edits, dates.
• Disclose your use of AI when it is required or expected by your audience.
• Consult a professional (an intellectual property attorney or solicitor, or the collective management organization in your field) for any commercial project or dispute.

If a term is unclear, our AI glossary explains it simply. To explore creative tools, start with our complete guide to generating images with AI.