On 20 October the Executive Board of NEMO, the Network of European Museum Organisations, published a statement after its Vilnius conference, Human after all: Museums in the wake of AI. More than 300 museum professionals from 40 countries took part. The statement’s first recommendation is simple: the sector should “develop clear policies, guidelines and a shared understanding of the ethical use of AI”, and “leadership is key” (NEMO).
I agree, and I’d go further. I developed AI policy for the V&A, helping one of the world’s leading cultural institutions think through responsible AI adoption. I’m not going to describe that document here; it belongs to the museum. But the work left me with a clear view: an AI policy isn’t an IT rulebook. For a museum, gallery or archive it’s a promise to the public about what you will and won’t do with the trust they give you.
Does a museum really need an AI policy?
Yes, and most already need one whether or not they have written it. Staff are using chatbots to draft labels, grant applications, social posts and emails today. Without a policy, every one of those people is making their own rules. NEMO’s statement puts the stakes plainly: museums are trusted, and “that same trust, however, makes museums a target”, with corrupted collection data and disinformation named as risks (NEMO). A policy is how you protect that trust on purpose rather than by luck.
What should a museum AI policy include?
Fewer pages than you think, and more decisions. In my view a useful policy answers six questions:
- What will we never do? Write your red lines first. For example: no personal or confidential data in unmanaged tools, and no AI-generated images presented as historical record.
- Which tools are approved? Name them. Staff follow a list far more readily than a principle.
- Who is responsible for an output? Always a named person, never the tool.
- What do we tell the public? Decide when you label AI-assisted interpretation, images, audio guides and chat.
- How do we treat collections and rights? Set rules for collection images, artists’ work, archives and community knowledge before anyone asks to train or fine-tune on them.
- When do we review it? Give it a date. The technology will move faster than your committee cycle.
Good examples are already public. North East Museums approved its AI policy on 30 September 2026 with a review date of September 2027. It rests on three rules, and the first is my favourite line in any policy this year: “You can’t say ‘AI did it’.” Its data rule bars personal or confidential information from any generative AI tool other than the managed one staff are given (North East Museums). That is the kind of sentence people remember on a busy Tuesday.
Should a policy ban AI or encourage it?
Neither, on its own. A policy that only bans things gets ignored, and one that only encourages gets people into trouble. An essay published by the American Alliance of Museums in August proposes three laws for museum AI governance in a fixed order: “Do No Harm. Preserve Trust. Create Value.” (American Alliance of Museums). I like the order. Protection comes first, but the third law is there for a reason. If AI can make your collection easier to search, your access work better or your small team less stretched, a good policy should make that easier, not harder.
What about AI agents?
This is the gap I see most often. Many policies were written for chatbots that answer questions. The newer tools take actions: they send emails, book things, publish and fill in forms. The Arts Marketing Association’s AI Policy Workbook, updated in June 2026, added a section on agentic AI, and its line is a good one to borrow: no agentic tool may take “consequential external actions”, including communications with audiences or funders, or any financial transaction, “without explicit human approval” (CultureHive). If your policy doesn’t mention agents yet, that’s the first edit to make.
What does the law require of museums using AI?
I’m not a lawyer, and you should take legal advice for your own situation, but two things are worth knowing. In the EU, the AI Act’s transparency rules have applied since 2 August 2026: deployers must disclose deepfakes and label AI-generated text published to inform the public on matters of public interest where there has been no human review or editorial control (European Commission). The Digital Omnibus left those duties in place and gave providers of existing generative tools until 2 December 2026 to add machine-readable marking (EY). If you have EU audiences, EU venues or EU partners, your labelling rules matter now.
In the UK, the picture is still moving. The government has said a broad copyright exception for AI training with an opt-out is “no longer the government’s preferred way forward” and that it will gather more evidence (GOV.UK). It has also set up work on labelling AI-generated content and on digital replicas (Hansard). For a museum, that uncertainty is a reason to write your own position on collections and rights now, not to wait.
How do you get staff to actually follow it?
Make it short, specific and kind. In my experience across the V&A, the BBC and Riot Games, the rules that stick are the ones people can repeat without opening the document. I’d aim for one page of commitments everyone signs up to, then practical guidance for each team. The AMA workbook suggests a similar three-layer structure: the policy, operational guidelines, and AI-specific additions to the policies you already have for data, HR, accessibility and publishing. Then run a short session with each team on their real tasks. A policy no one has practised is just a PDF.
Where should a museum start this week?
Ask three teams what they already use AI for, and write the answers down. Decide your red lines. Pick your approved tools. Add a paragraph on agents and one on labelling. Put a review date on it. You’ll have a working draft before the end of the month, and a better one than most institutions have today. The museums that get this right won’t be the ones with the longest policies. They’ll be the ones the public still trusts in five years.
Frequently asked questions
Does a museum need an AI policy?
Yes. Staff in most museums, galleries and archives already use generative AI tools for drafting and research, so the institution needs clear rules on data, approved tools, responsibility and public disclosure. NEMO’s Executive Board recommended on 20 October 2026 that the sector develop clear AI policies and guidelines.
What should a museum AI policy include?
RE/VERSE suggests six things: red lines (what the institution will never do), a list of approved tools, a named person responsible for every output, rules for labelling AI use to the public, rules for collections, artists’ rights and archives, and a fixed review date.
Are there example AI policies for museums?
Yes. North East Museums published an AI policy approved on 30 September 2026, and the Arts Marketing Association publishes a free AI Policy Workbook on CultureHive (version 3, June 2026) that cultural organisations can adapt.
How should a museum AI policy deal with AI agents?
It should limit what autonomous tools can do on the institution’s behalf. The AMA workbook says no agentic AI tool should take consequential external actions, such as communicating with audiences or funders or making payments, without explicit human approval.
Does the EU AI Act affect museums?
It can. The AI Act’s transparency rules under Article 50 have applied since 2 August 2026, including duties for deployers to disclose deepfakes and label some AI-generated public-interest text. Institutions with EU audiences, venues or partners should take legal advice on how this applies to them.
Who is RE/VERSE?
RE/VERSE is an independent creative consultancy working across culture, games, fashion and emerging technology.
About RE/VERSE
RE/VERSE is an independent creative consultancy working across culture, games, fashion and emerging technology.
Sources
- NEMO: NEMO Executive Board publishes statement on museums and AI (20 Oct 2026)
- North East Museums: Artificial Intelligence Policy and Guidance (approved 30 Sep 2026)
- CultureHive / Arts Marketing Association: AI Policy Workbook, version 3 (June 2026)
- American Alliance of Museums: The Three Laws of AI Governance (31 Aug 2026)
- European Commission: Quick facts, transparency rules for AI systems (Article 50, AI Act)
- EY: Regulation (EU) 2026/1744, Digital Omnibus on AI
- GOV.UK: Report on Copyright and Artificial Intelligence (2026)
- Hansard: Copyright and AI, written statement (18 Mar 2026)
- NATAS: Sports Emmy Awards winners release, Outstanding Esports Coverage (Riot Games, League of Legends Worlds)