AI Agents for Museums and Cultural Institutions: What the First UNESCO-ICOM Global Survey Found

The First Global Count: 57% of Museums Are Already Using AI

UNESCO and the International Council of Museums (ICOM) ran the first-ever global survey on AI use in museums, presenting the findings at the Global Forum on the Ethics of Artificial Intelligence in Riyadh in September 2026. They surveyed more than 400 museums across 90 countries, and the topline finding is that adoption arrived faster than anyone tracking the sector expected: 57% of responding museums are already using AI in some form. That's not a pilot program at a handful of flagship institutions — it's a majority, spread across a sector not usually associated with moving fast on new technology.

But the survey's second finding is the one that matters more: 55% of those same museums have no internal AI policy, strategy, or guidelines of any kind. Adoption has outrun governance almost everywhere at once.

Where the Work Is Actually Landing

The survey breaks down exactly what museums are using AI for, and the ranking is telling: administration leads at 70%, followed by communication at 62%, and research and documentation at 43%, with exhibition development and visitor engagement further behind. That ordering matters because it's the opposite of the stereotype — the loudest public conversation about AI in museums is usually about generative exhibits or AI-written interpretive text, but the survey shows the actual center of gravity is internal: grant reports, scheduling, correspondence, cataloguing, and the unglamorous administrative load that every museum carries regardless of size.

ICOM's Director-General, Medea S. Ekner, framed the finding directly: "Museums around the world are already experimenting with AI, while navigating important questions around ethics, trust and professional responsibility." UNESCO Assistant Director-General for Culture Nayef Al-Fayez put the gap in institutional terms: "the results of the global survey point to a sector at an important stage of AI adoption: experimentation is advancing faster than institutional capacity and governance."

Training Hasn't Caught Up to Usage

Only 17% of museums report providing any AI training to staff — meaning the large majority of the 57% already using AI got there through individual staff experimentation rather than an institutional rollout with guidance attached. The survey's respondents named their own top concerns without being prompted much: accuracy, copyright and intellectual property, data protection, transparency, and bias. Their most common request wasn't for more AI tools; it was for practical guidance, training, stronger data governance, and a way to share what's working with peer institutions — the exact infrastructure that 55% of them currently lack.

The survey also documents real variation in how differently-resourced institutions are approaching this, not just a single adoption curve: community-led AI for safeguarding cultural knowledge at the Maasai Heritage Museum in Kenya, AI-supported visitor engagement at the Museum of Tomorrow in Brazil, and narrower applications for collections research and accessibility elsewhere. Adoption is real, but it isn't uniform, and it isn't following a single playbook.

The Public Isn't as Comfortable as the Adoption Numbers Suggest

A separate data point, cited by the American Alliance of Museums from the 2026 Annual Survey of Museum-Goers, complicates the picture further: 70% of the general public want museums to use no AI at all when it comes to developing exhibitions, and 43% felt museums shouldn't even use AI to write emails or website text. That's a meaningfully more skeptical public than the museums' own adoption numbers might suggest, and it lines up with the backlash several institutions — including the British Museum — have already faced over AI-generated imagery in their public-facing content this year.

That gap between what museums are doing internally (administration, largely invisible to visitors) and what the public would tolerate seeing (almost nothing, especially in exhibitions) is arguably the more useful finding for any museum leader reading the UNESCO-ICOM data. The safest, most-adopted use case — administrative work — is also the one visitors never see and therefore never object to. The riskiest use case, reputationally, is the one closest to the museum's actual public mission.

What This Means for Museums and Cultural Institutions Going Forward

The UNESCO-ICOM survey makes clear that the "should we use AI" conversation is already over for most of the sector — 57% adoption settled that. The conversation that's actually still open is governance: only 17% of museums have trained their staff, and 55% have no policy at all guiding what "using AI" is even supposed to mean at that institution. For a sector built on public trust, that's the riskier gap of the two. The institutions best positioned going forward look less like the ones racing to add the most visible AI feature to an exhibit, and more like the ones that write down a policy, train the staff already experimenting on their own, and keep AI concentrated where the survey shows it's already working quietly: administration, documentation, and communication — not the exhibition floor.