Artificial intelligence in exhibitions and institutions

Reviewed as of 1 October 2026. If you spot an error, write to me.

This guide is for people who work in a museum, a gallery, a cultural centre or a foundation, and for artists who exhibit works made with artificial intelligence. It starts from the premise that AI is already in use and separates two questions: which legal obligations may apply and what an institution must decide in order to use it responsibly.

Adoption faster than the rules

In September 2026, UNESCO and the International Council of Museums published the results of their first global survey on the use of AI in museums. A total of 466 responses were received from ninety countries and 403 were retained for the analysis. The results describe the institutions that responded; they are not a representative estimate of all museums.

Of the respondents, 57% said they use AI and 55% said they have no internal policy, strategy or guideline. Among the museums that used it, the uses cited included administration, at 70%; communication, at 62%; and research and documentation, at 43%. Only 17% of respondents said they offer AI training to their staff. The concerns highlighted were accuracy, copyright and intellectual property, data protection, transparency and bias.

The questionnaire was launched while ICOM's 2004 Code of Ethics was still in force. During the survey period, the ICOM General Assembly adopted a revised code on 25 June 2026. The new text expressly recognises the impact of digital technologies and sets out professional and governance principles, but it does not replace an operational policy on AI.

Transparency from August 2026

Article 50 of the Artificial Intelligence Act has applied since 2 August 2026. The obligations depend on the institution's role: developing or marketing a system is not the same as using someone else's.

Systems that converse with the public

Whoever provides a system intended to interact directly with people must design it so that they know they are interacting with an AI, unless this is obvious to a reasonably well-informed, observant and circumspect person in the specific circumstances. If an institution develops or puts into service its own assistant, it must determine what role it occupies and what obligations fall to it.

Technical marking of synthetic content

Providers of systems that generate synthetic audio, image, video or text must ensure that the outputs are marked in a machine-readable format and can be detected as artificially generated or manipulated, as far as is technically and reasonably feasible. This obligation falls on the provider of the system, not automatically on the museum that uses a third party's tool.

Deep fakes and artistic works

Whoever uses a system to generate or manipulate an image, audio or video that constitutes a deep fake must disclose that the content has been artificially generated or manipulated. Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the obligation is limited to disclosing the existence of generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work.

This exception shapes how information is given; it does not remove the obligation. Nor does it turn every work made with AI into a deep fake: the legal category requires the content to resemble existing people, objects, places, entities or events and to appear authentic or truthful.

Texts of public interest

Whoever uses AI to generate or manipulate text published in order to inform the public on matters of public interest must disclose that it has been artificially generated or manipulated. The obligation does not apply when the content has undergone human review or editorial control and a natural or legal person takes editorial responsibility for its publication.

A minimum policy

The law does not exhaust the decisions an institution must make. A short policy can answer these questions:

What it is used for. Translation, drafts, public enquiries, research, documentation, cataloguing, accessibility, communication or image generation.
Who is responsible. Who reviews each output and who authorises a new use.
What is checked. Dates, attributions, provenance, quotations, rights, personal data and any factual claim.
What is communicated to the public. When people are informed, where, and in how much detail about the role of AI.
What data is handed over. Images from the collections, visitor information, internal documents, sensitive data or protected content.
What the contract allows. Data retention, training, confidentiality, security, location of processing and the possibility of export.
What is kept. Prompts, versions, human reviews, sources and the decisions needed to reconstruct how a piece of content was produced.
How to complain. What happens if someone spots an error, a bias, a false attribution or an unauthorised use.

Works, images and rights

That a work is in the collection does not mean the institution holds its exploitation rights. Before using images or texts in a system, check ownership, licence, purpose and the provider's terms.

Works in the public domain are no longer subject to the author's exploitation rights, but a particular image may carry other rights or be conditioned by access to the archive, the service contract or the originality of the reproduction. The guide Images of public domain works explains these differences.

If the institution publishes its own content online, it can reserve its rights against text and data mining by appropriate, machine-readable means. The guide How to reserve your rights against data mining explains the scope and limits of those measures.

If you exhibit a work made with AI

There is no single formula for the wall label. Describe with sufficient precision:

which system or systems were involved;
what materials the artist supplied;
what decisions, selection and editing the artist carried out;
whether the work changes during the exhibition;
whether it uses data, images, voices or faces of third parties;
which components must be preserved.

A technical description does not by itself settle legal authorship, nor does it require publishing secrets or complete instructions. It should allow the process to be understood without attributing to the system an agency it does not have and without concealing a relevant intervention.

The exhibition contract should clarify the reproduction of the work, gallery documentation, promotional uses, the preservation of files and the possible use of the images by AI services. Do not take a training permission for granted merely because the institution may photograph or communicate the work.

On the open conversation

This guide summarises general criteria. Article 50 already has Commission guidelines, but its application depends on the system, the content and the legal role of each party. If you work in an institution that has adopted an AI policy and spot an omission, write to me.

Sources

Discover more from Juan A. Esteban

Subscribe to receive the latest posts by email.