Museums Are Adopting AI Faster Than They Write the Rules: Lessons on AI Governance for Cultural Venues from TTXC and the UNESCO-ICOM Global Survey

57% of museums worldwide already use AI, yet 55% have no internal rules. Before bringing AI into exhibits, set usage rules and run a risk assessment first.

In October 2026, the TTXC Taiwan Technology X Culture Expo, organized by Taiwan's Ministry of Culture in Kaohsiung, drew 987,000 visits over 19 days under the theme "AI Dimension: Symbiotic Era." Public appetite for AI-driven interactive experiences is clear. Meanwhile, the UNESCO-ICOM global survey on AI use in museums, released in September, shows that most institutions are still at the stage of individual staff experimenting on their own, with governance and training lagging well behind. This article pulls together these two official sources, compares them with Taiwan's AI Basic Act, and outlines what museums, curatorial teams and their technology partners can do now.

1. Visitors are on board: nearly a million visits in 19 days at TTXC

The takeaway: AI interactive exhibits have moved from novelty experiments to mainstream content people will queue for. The question for institutions is no longer whether to do it, but how to do it reliably.

According to the Ministry of Culture's press release of October 29, 2026, this year's TTXC was co-organized with the Kaohsiung City Government, the Taiwan Creative Content Agency, the Ministry of Digital Affairs, the Ministry of Economic Affairs, the National Science and Technology Council and others. It featured 7 themed concept exhibitions, 6 future-facing performances, more than 100 industry and matchmaking events and 25 forums, and recorded 987,000 visits between October 8 and 26.

The most popular themed exhibition, "Yes, AI Do. Welcome to the 2050 Emotion Exchange," guided visitors through 5 experience stages to generate their own virtual partner. It attracted more than 25,000 visits and produced over 15,000 virtual couples. The XR DREAMLAND section presented 39 Taiwanese and international XR works, 11 of them Taiwanese originals or international co-productions. The Ministry of Digital Affairs' DigiWave exhibition applied real-time 4DGS digital twin creation in an exhibition setting, and the Ministry of Economic Affairs joined for the first time, showcasing 6 technologies including generative AI art and video and photorealistic 3D model construction.

An Executive Yuan press release on the same day noted that AI and the cultural and creative industries are both among the Executive Yuan's "13 strategic industries," and asked the Ministry of Culture to work with the Ministry of Digital Affairs, the Ministry of Economic Affairs and other agencies on policy coordination and resource integration. In other words, culture-tech is also a policy priority, and opportunities for venues and events across Taiwan to adopt AI will only grow.

2. The global museum gap: many users, few rules

The takeaway: museums are adopting AI faster than they are building internal governance and staff training, and that gap is currently the largest source of risk.

The first joint UNESCO-ICOM global survey on the use of AI in museums drew responses from more than 400 museums in 90 countries. The results were released on September 17. Key figures:

57% of responding museums already use AI
55% have no internal AI policy, strategy or guidelines
Only 17% provide AI training to staff
The largest area of use is administration (70%), followed by communication (62%) and research and documentation (43%)

As the survey's release put it, experimentation is moving faster than institutional capacity and governance. AI use remains mostly exploratory or driven by individual staff rather than institution-wide. Respondents' leading concerns were accuracy, copyright and data protection, and the support they most wanted was practical guidance and training on the technical and ethical use of AI, along with stronger data governance.

The survey also points out that access to AI technologies and the resources to implement them remains uneven, with gaps in digital infrastructure, funding and technical capacity affecting outcomes. For small and mid-sized museums on limited budgets, this means it is even more important to define scope, rules and responsibilities at the outset, rather than spending scarce resources on fixes after the fact.

3. Taiwan's legal framework: public institutions already have clear obligations

The takeaway: Taiwan's AI Basic Act is in force. For public museums and government agencies using AI, risk assessment and internal control rules are legal requirements, not optional extras.

The AI Basic Act was promulgated by the President on January 14, 2026 and took effect the same day. It has 20 articles. Three are most directly relevant to cultural venues.

Article 19: government use of AI requires a risk assessment

When the government uses AI to carry out its work or provide services, it must conduct a risk assessment and plan response measures, and set usage rules or internal control mechanisms based on the nature of the work. Public museums, art museums and cultural centers introducing AI guides, AI customer service or generative interactive exhibits should build both into their plans.

Article 4: seven basic principles

The Act lists sustainable development and well-being, human autonomy, privacy protection and data governance, cybersecurity and safety, transparency and explainability, fairness and non-discrimination, and accountability. The requirement that AI outputs be appropriately disclosed or labeled is especially important in galleries: visitors need to know when the commentary, images or conversation in front of them were generated by AI.

Article 14: build personal data protection into the design

The Act requires avoiding unnecessary collection, processing or use of personal data in AI development and use, and promotes privacy by default and by design. Interactive exhibits often take photos, record audio or ask visitors to type text. Whether to keep that data, for how long and where it is stored should all be decided at the design stage.

4. From gallery to back office: three risks to address first

The takeaway: accuracy, copyright and visitor data map directly onto museums' top concerns worldwide, and they are the items that most need to be written into specifications before deployment.

Accuracy of interpretive content

Generative AI guides or conversational exhibits can produce statements that conflict with the museum's own research. A practical approach is to have the AI answer only from collection data and texts the museum has approved (a RAG architecture), with a standard response for when no answer is found, rather than letting the model improvise.

Copyright in collection images and texts

Before giving collection images, research texts or partner artists' works to an external AI service, confirm the scope of licensing and whether the provider will use the data to train its models. Contracts should spell out data use, retention periods and deletion procedures.

Visitor data and cybersecurity

Once interactive installations, ticketing and membership systems are connected to AI, visitor data flows become more complex. Access permissions on devices and in the cloud, log retention and incident reporting procedures should be accepted and signed off together with the exhibit itself.

5. Recommendations for organizations

The takeaway: starting with rules and small pilots is safer than buying a large AI exhibit all at once. Here are 5 things museums, curatorial teams and technology partners can start on right away.

Take stock: list the AI tools and uses already present in your organization, including generative AI services staff use on their own. The UNESCO survey shows most AI use starts with individual staff; you can only manage what you can see.
Write a one-page AI usage policy: specify which data must not be entered into external services, which outputs need human review, and how public-facing content generated by AI is labeled. Public institutions can map this directly to the risk assessment and internal control requirements of Article 19 of the AI Basic Act.
Pilot in low-risk back-office work: the survey shows administration is the most common area of use. Start with internal processes such as meeting notes, document organization and first drafts of multilingual translation, then extend to visitor-facing exhibits once you have experience.
Write governance clauses into procurement specifications: requirements for interactive exhibits and AI guides should include answering only from museum-approved content, retention periods for visitor data, no leakage of models or data, and deadlines for reporting security incidents. If you need help turning these conditions into technical specifications, work with a systems vendor that holds an information security management certification (such as ISO/IEC 27001). JIE HENG's cultural venue and interactive technology services also cover this step.
Provide basic AI training: only 17% of museums train their staff, which makes this the easiest gap to close. Training need not be complicated; focus on situations where AI makes mistakes, your internal rules, and how to report problems.

AI exhibits can draw visitors in, but what keeps visitors trusting an institution is clear rules behind the scenes. Setting those rules while deployments are still small is the cheapest time to do it.