Illustration of Juan Esteban’s runic monogram rendered in a three-dimensional architectural style within a frame, set against a light background.

How to disclose the use of artificial intelligence in an artwork, sale or open call

‘Made with AI’ can conceal very different processes: searching, translating, retouching, generating, classifying or fine-tuning a model. A useful disclosure identifies the system, the stage, the materials and the human decisions without attributing more certainty than can be shown.

Five sources of obligation

Distinguish between:

  • a legal rule;
  • the terms of an open call;
  • a contract or a responsible declaration;
  • the platform terms;
  • a voluntary disclosure by the artist.

Complying with one does not mean complying with the others. Read the current text before submitting or selling.

Which system was used

Record the provider, product, model, version, whether it ran locally or through an API, and the date. If you do not know the version, say so. Keep the base model, your own model fine-tuning and any auxiliary tools separate.

At what stage it intervened

State whether it was used for research, sketching, text, image, audio, code, selection, editing, translation, restoration or presentation. Do not equate a spelling correction with substantial generation.

Materials and data

Declare which materials are your own, licensed, in the public domain or supplied by third parties, where this is relevant. If you carried out model fine-tuning, identify the dataset without publishing reserved information. The guide Training an AI model on your own work: rights, data and documentation covers the dataset manifest and the documentation of the fine-tuning. To reserve rights over your own materials, see How to reserve your rights against data mining.

Human contribution

Describe verifiable decisions: concept, selection, composition, editing, programming, discarding, installation and later transformation. Avoid empty formulas such as ‘supervised by a person’. A disclosure does not by itself settle whether the artwork is protectable or who its author is. That question is dealt with in Artificial intelligence and copyright in Europe.

Article 50 of the AI Act

Article 50 of Regulation (EU) 2024/1689 brings together the transparency obligations and applies from 2 August 2026. It divides the obligations between two roles, the provider of the system and the deployer:

  • Systems intended to interact directly with people must inform them that they are dealing with an AI, unless this is obvious (paragraph 1, providers).
  • Systems that generate synthetic audio, image, video or text must mark their outputs in a machine-readable format, so that they can be detected as artificially generated or manipulated (paragraph 2, providers).
  • Deployers of an emotion recognition or biometric categorisation system must inform the people exposed to it (paragraph 3).
  • Deployers of a system that generates or manipulates image, audio or video constituting a deepfake must disclose that the content is artificial. They must also disclose it when they publish text generated or manipulated with AI to inform the public on matters of public interest, unless the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility (paragraph 4).
  • The information must be clear and distinguishable, at the latest at the time of the first interaction or exposure (paragraph 5).
  • The article does not replace other transparency obligations under Union or national law (paragraph 6).
  • The Commission encourages codes of practice on the detection, marking and labelling of generated content (paragraph 7).

For evidently artistic, creative, satirical, fictional or analogous works that include a deepfake, the text limits the obligation to disclosing its existence in an appropriate manner that does not hamper the display or enjoyment of the artwork. The exception shapes the form of the disclosure. It does not remove the duty to inform within its scope.

Not all generated content is a deepfake. Consider whether it imitates existing people, objects, places, entities or events in a way that could lead someone to believe it is authentic.

This guide does not determine whether an artist is a provider or a deployer. That role depends on how the system is used. Check your case against the text of the AI Act.

Transitional period for marking

Regulation (EU) 2026/1744 of 8 July 2026 amends the AI Act. Its recital 38 provides for a four-month transitional period for the machine-readable marking under Article 50(2), for providers of generative systems that had already placed them on the market before 2 August 2026. The notes of Cuatrecasas and Garrigues place its end on 2 December 2026; that date has not been checked directly against the operative articles of the Regulation. It concerns the technical marking carried out by providers. It does not alter the disclosure obligations in paragraph 4.

Retouching and assisted editing

Article 50(2) does not apply to systems that perform an assistive function for standard editing, or to those that do not substantially alter the input data provided by the deployer or its semantics. That exception concerns the technical marking that falls to the provider. It does not mean that every use of an assisted function requires disclosure, nor that it is always exempt.

To decide what to disclose, separate minor assistance (correcting, cropping, adjusting colour) from substantial generation or manipulation. When in doubt, describe the actual stage in your disclosure.

Open calls

Read the definition, the prohibitions, the documentation, the period and the consequences; the guide How to read the rules of an open call suggests an order of reading. Answer the form exactly. If the terms of the open call require a responsible or sworn declaration, do not add reassuring interpretations that alter its content.

The terms of each open call may admit, require disclosure of, or exclude certain uses of artificial intelligence. No uniform rule applicable to all Spanish artistic open calls has been located.

Sale and description

The sales listing should state clearly what is on offer: file, print, model, token, certificate or licence. Disclose the use of AI when concealing it could mislead as to the process, authorship, technique or nature of the item. Article 50 does not impose a specific label for this case; the requirement to describe accurately depends on consumer and competition law, which this guide does not analyse.

Do not promise exclusivity or rights over the outputs without reviewing the model, the materials and the contract.

Platform and social media

Check each platform’s terms on labelling, synthetic content, rights and metadata. They are contractual conditions of their own, distinct from legislation, and they may change. Keep a dated capture of the terms and keep the disclosure outside the platform, because labels can change or disappear.

The European Commission has published a set of icons for labelling AI-generated content. Their use is optional.

Evidence of the process

Keep the relevant prompts, seeds, inputs, outputs, versions, metadata, intermediate files, discarded material, code, date and captures of the applicable terms. The documentation does not create authorship, but it allows the extent of the intervention to be checked. The guide Artificial intelligence in exhibitions and institutions covers the policies of the institutions that receive such artworks.

Short model of a disclosure

Adapt these fields to the recipient:

  • System used: [system, model and version where known].
  • Stage: [stage at which it intervened].
  • Materials or data processed: [description without revealing reserved information].
  • Contribution of the author: [specific decisions and transformations].
  • Generated or manipulated content that remains in the artwork: [extent].
  • Date of review: [date].

Do not use the model if the terms of an open call require a different form.

Checklist

  • The source of the obligation has been identified.
  • System, version and date are recorded.
  • The actual stage of intervention is described.
  • Materials and rights have been reviewed.
  • The human contribution is explained with facts.
  • It has been considered whether there is a deepfake or a text on a matter of public interest.
  • The artistic exception was not treated as a dispensation.
  • Minor editing assistance was not confused with substantial generation.
  • Open-call terms, contract and platform terms were reviewed separately.
  • The sale does not mislead as to technique or nature.
  • Evidence of the process is kept.

Further reading

Sources

Status: Verified. Reviewed: 11 October 2026. General information, not an interpretation of specific terms or a specific contract. If you find a mistake, write to me.


Discover more from Juan A. Esteban

Subscribe to receive the latest posts by email.

Español English (UK)