Museums’ AI governance gap extends beyond writing a policy

Museums are experimenting with artificial intelligence faster than they are building the policies and skills needed to govern it. The first UNESCO–ICOM global survey on AI in museums found that 57% of more than 400 responding institutions across 90 countries already use AI, while 55% have no internal policy, strategy or guideline.[1][2][3]

The result describes an adoption gap, not simply a technology gap. Many museums are using accessible tools through individual staff members before the institution has decided which data may be uploaded, how generated material should be labelled, who checks accuracy or what happens when a vendor changes its terms.[1][2][3]

Use is broad, but institutional capacity is thin

ICOM’s summary says adoption remains largely exploratory or staff-led rather than institution-wide. Respondents identified accuracy, copyright, intellectual property, data protection, transparency and bias among their leading concerns. They asked for practical guidance, stronger data governance and training in both technical and ethical use.[2]

Museum errors have a long half-life

An inaccurate marketing draft can be corrected. An error introduced into a collection record may propagate into exhibitions, research databases, educational material and other institutions’ linked data. The same is true of provenance and attribution. A model that fills a metadata gap with a plausible invention can make a collection appear more complete while quietly making it less trustworthy.[1][2][3]

That makes source traceability unusually important. Museums need to distinguish generated description from catalogued fact, preserve the evidence behind revisions and ensure that staff can reverse automated changes. The institution’s authority depends not on never making mistakes, but on being able to show how a claim entered the record and how it was checked.[1][2][3]

Trust is becoming a security question

After a conference in Vilnius involving more than 300 museum professionals from 40 countries, the Network of European Museum Organisations called for clear policies, shared ethical understanding and a stronger role for museums in European AI policy. NEMO’s statement links AI to cyberattacks, disinformation and manipulation of digital heritage.[3]

The connection is direct. Museums hold politically and culturally sensitive records. Corrupted collection data, fabricated provenance or altered digital reproductions can be used to contest ownership, erase communities or manufacture historical narratives. Institutions that are trusted by the public are valuable targets precisely because their endorsement carries weight.[1][2][3]

Governance must be proportional to the use

Not every museum needs a large AI office. A practical policy can begin with a use register, approved tools, data-handling rules, human review and disclosure standards. Higher-risk uses—collection cataloguing, provenance research, facial recognition, rights decisions or public historical interpretation—need stronger evidence and approval than drafting an internal meeting summary.[1][2][3]

Training should also be role-specific. A communications team needs to recognise fabricated quotations and image-rights problems. Registrars and curators need to understand hallucinated metadata and provenance uncertainty. IT and security teams need vendor, access and logging controls. Leaders need a way to decide when an efficiency gain is not worth the institutional risk.[1][2][3]

Museums can shape AI, not only adopt it

The sector’s value to AI policy is not confined to collections. Museums have long experience with contested interpretation, provenance, public explanation and stewardship across generations. Those disciplines can inform how AI systems document evidence, represent uncertainty and handle culturally sensitive material.[1][2][3]

The survey therefore captures a choice. Museums can treat AI as an informal productivity layer and attempt to add governance after problems emerge, or they can use their own professional traditions to define responsible adoption. The technology is arriving quickly. The authority to say what is known, where it came from and why the public should trust it has to be protected deliberately.[1][2][3]

Report an error or suggest a correction