Skip to content
Policy · 9 min read

Do AI-made books need a label in the EU? AI Act Article 50 explained (2026)

Short answer

Not as a general rule. Article 50 of the EU AI Act, in application since August 2, 2026, makes AI providers mark outputs, and makes professional users label deepfakes and AI text published to inform the public on matters of public interest. How that applies to a self-published novel or picture book is not settled yet.

Key numbers

  • In application since August 2, 2026; providers of tools already on the market mark outputs by December 2, 2026
  • 4 obligations: chatbots disclosed, outputs marked, emotion recognition disclosed, deepfakes and public-interest text labelled
  • No label for text with human review or editorial control and an editorially responsible person
  • Code of Practice on marking and labelling: final on June 10, 2026, voluntary

By kdpbook.io editorialPublished ✓ Sources read on

The question comes up because the EU’s transparency rules started on a fixed date this year and say “text” and “image” without saying “book”. Below is what the Regulation and the Commission actually say, quoted from EUR-Lex and the Commission’s own pages as we read them on November 22, 2026, and the points no authority has settled for publishing. This is information, not legal advice.

What does Article 50 say?

Article 50 of Regulation (EU) 2024/1689, the AI Act, sets four transparency obligations. It is not among the provisions Article 113 dates differently, so it applies from August 2, 2026.

ParagraphWhoObligationRelevant to books?
50(1)ProvidersPeople must be told they are interacting with an AI system (a chatbot, for example), unless it is obviousRarely: it concerns interactive systems
50(2)Providers of systems generating audio, image, video or textOutputs “marked in a machine-readable format and detectable as artificially generated or manipulated”Indirectly: it binds the AI tool, not the author
50(3)Deployers of emotion recognition or biometric categorisationInform the people exposed to itNo
50(4), first partDeployersDisclose that deepfake image, audio or video content was artificially generated or manipulated; for an evidently artistic, creative, satirical or fictional work, a disclosure “that does not hamper the display or enjoyment of the work”For realistic AI pictures of real people, places or events
50(4), second partDeployersDisclose AI text “published with the purpose of informing the public on matters of public interest”, unless it had human review or editorial control and someone holds editorial responsibilityFor some nonfiction; unclear for most books

Article 50(5) asks for the information “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure”. According to the Commission’s Article 50 FAQ, fines for these obligations can reach 15 million euros or 3% of total worldwide turnover for the preceding financial year.

Who does Article 50 apply to?

  • Providers develop an AI system or model and place it on the market under their own name: the companies behind the text and image tools.
  • Deployers are anyone “using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity” (Article 3(4)). The Commission’s FAQ says an activity through which a person “gain[s] an economic benefit on a regular basis” is professional. An author who regularly sells books made with AI tools would, on that reading, be a deployer.
  • Outside the EU: Article 2(1)(c) covers providers and deployers established in a third country “where the output produced by the AI system is used in the Union”.

Does a book with AI-written text need a label?

Only text that meets the second part of Article 50(4) must be labelled. The Commission’s FAQ reads it as three criteria together: the text is published, informs the public, and is on matters of public interest, which it lists as politics and democratic processes, public administration, justice, fundamental rights, public security, public health, environmental protection, consumer safety and economic, financial, political, scientific or cultural developments that may be the subject of public debate.

  • A novel, a children’s story, a coloring or puzzle book: not text published to inform the public, on the face of these criteria. No authority has said so for books specifically.
  • Nonfiction on health, politics, finance or science sold in the EU: the criteria can fit, so the exemption matters.
  • The exemption: text that “has undergone a process of human review or editorial control” with “a natural or legal person” holding “editorial responsibility for the publication”. The FAQ says superficial, formal or procedural checks, such as spell-checking or grammatical correction, don’t count. Its examples of real review (an editor-in-chief, academic peer review) come from journalism and science; how they map to a self-published author is not settled.

Do AI pictures in a book need a label?

Only deepfakes. Article 3(60) defines a deep fake as AI-generated or manipulated image, audio or video content “that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”. The Commission’s FAQ makes the three criteria cumulative and says the context and the expected audience count: content the audience doesn’t expect to be authentic may not falsely appear so. An imaginary cartoon dragon resembles nothing that exists; a realistic AI photo of a real person or event in a nonfiction book can.

Even then, a book is usually an artistic, creative or fictional work, for which Article 50(4) limits the duty to disclosing that such content exists, in a way that doesn’t spoil the work. No guidance we found says where in a book that disclosure belongs.

Who marks AI output, and from when?

  • From August 2, 2026: providers of generative systems mark outputs as AI-generated in a machine-readable, detectable way (Article 50(2)).
  • Until December 2, 2026: Article 111(4), added by Regulation (EU) 2026/1744 of July 8, 2026, gives providers of systems placed on the market before August 2, 2026 until December 2, 2026 to comply with Article 50(2).
  • Code of Practice: the Commission published the final Code of Practice on marking and labelling AI-generated content on June 10, 2026. It is voluntary.
  • Older content: the Commission’s FAQ says content generated before August 2, 2026 doesn’t need to be labelled retroactively.

Marking is the tool maker’s duty. It is separate from the labels of Article 50(4): a provider’s watermark doesn’t replace a label a deployer owes, and removing one doesn’t remove the other.

Is KDP’s AI disclosure the same thing?

No. KDP’s Content Guidelines require you to inform KDP of AI-generated text, images or translations when you publish or republish a book, wherever you sell. That answer goes to Amazon, under Amazon’s rules; Article 50 is EU law about what readers are told. A book can need one, both or neither. The KDP AI disclosure guide covers the form, and the copyright guide covers who owns AI text and pictures.

What is still unsettled?

  • Whether selling a book to EU readers is “use in the Union” of the AI system’s output for an author based elsewhere.
  • When a book counts as text “informing the public on matters of public interest”: the Act and the FAQ don’t mention books.
  • How much review makes a self-published author’s editing “human review or editorial control”.
  • Where a disclosure in a book should sit to be “clear and distinguishable” at first exposure without hampering the work.
  • How national market surveillance authorities will apply all of this: none had published book-specific guidance that we found on November 22, 2026.

kdpbook’s own approach to AI content is on our AI content policy page. For a decision about your book, ask a lawyer who practises EU law.

Questions

Does the EU AI Act require a label on every AI-generated book?

No such general rule is in the text. Article 50, in application since August 2, 2026, requires labels in listed cases: deepfake images, audio or video, and AI text published to inform the public on matters of public interest. Separately, the providers of AI tools must mark their outputs in a machine-readable way. How these cases apply to a given book is for national authorities and courts to settle.

Does Article 50 apply to authors outside the EU?

It can. Article 2(1)(c) covers providers and deployers established outside the Union “where the output produced by the AI system is used in the Union”. Whether a book sold to EU readers counts as such use is one of the open questions; the Act’s text doesn’t mention books.

Is a self-publisher a ‘deployer’ under the AI Act?

The Act defines a deployer as anyone using an AI system under its authority, except in a personal non-professional activity. The Commission’s FAQ adds that an activity through which a person gains an economic benefit on a regular basis is professional. By that reading, an author who regularly sells books made with AI tools would be a deployer; no court has ruled on it yet.

Does editing AI text remove the label requirement?

For AI text on matters of public interest, Article 50(4) lifts the obligation when the text has undergone human review or editorial control and a person holds editorial responsibility for the publication. The Commission’s FAQ says spell-checking or grammatical correction doesn’t count as review.

When do AI tools have to watermark their output in the EU?

From August 2, 2026 for tools placed on the market from that date. Regulation (EU) 2026/1744 gave providers of tools already on the market before then until December 2, 2026 to comply with Article 50(2).

Make it with kdpbook

Let kdpbook handle the details.

Describe your book in one sentence and get KDP-ready files, with the rules from this guide already applied.

2,000 free credits · no card required