GLAMorous AI TL;DR — July 2026

Welcome to July's edition. This month we're exploring a question the sector is asking but rarely answering out loud: what happens to human expertise when AI systems become faster, cheaper, and more consistent at tasks curators have always done?

The answer matters. Because the conversation we're having in public (about models, guidelines, and technical readiness) is not the conversation happening in staff rooms across GLAM institutions.

🌟 Featured Reads

AI4LAM – Artificial Intelligence for Libraries, Archives and Museums

AI4LAM formalised this year as an official membership organisation, now hosted by the National Library of Norway. The Getty and Stony Brook University Libraries are among founding members. Fantastic Futures 2026 (Washington DC, September 14-18) will be the largest gathering of GLAM-AI practitioners to date.

What matters: this signals that AI adoption in GLAM is moving from experimentation to infrastructure. The question is no longer whether to engage with AI. It is how to do so without losing institutional identity and curatorial authority.

Read more →

🏛️ When the Model Becomes the Author

Wasielewski – AI and Curation: Digitization, Museums and Digital Art History

The curator's role has shifted over two decades. They are no longer neutral custodians but "auteurs": individuals whose subjective position, argument, and point of view are central to the profession. Now generative AI arrives with the capacity to make recommendations at scale, suggest connections, surface patterns, and generate interpretations faster than any human team.

The problem: if automation outsources curatorial judgment to an algorithm, what authority remains for the curator? And if curators become overseers of automated systems rather than authors of interpretation, does that erode the very thing that makes museums culturally significant?

The paper does not claim to know the answer. But it names the stakes.

Read more →

📊 What People Are Saying

Arantes – Algorithmic Curation and the Future of Museum Practice

Algorithms are curators. They decide what you see first, what gets recommended, what is grouped together, what remains invisible. Making these algorithms visible, editable, and contestable—building curatorial oversight into automated systems—becomes essential heritage work.

But here's what makes this uncomfortable: most institutions have not built that oversight. When recommendations appear without explanation, no one knows whether they reflect curatorial intention, training data bias, or commercial incentives.

Read more →

Mozilla Data Collective – Cultural Heritage and AI: How Institutions Can Reclaim Control of Their Data

A concrete model: Mozilla's Data Collective enables GLAM institutions to become intentional participants in the AI data economy, curating multicultural and multilingual datasets while retaining complete control over use. Institutions retain sovereignty. This is sovereignty in practice, not just principle.

Read more →

🚨 What People Are NOT Saying

The Staff Anxiety We Are Not Naming

No GLAM journal has yet published on what is happening in staff rooms. But the Canadian War Museum's Paul Durand said it plainly in December: institutions worry about "job obsolescence"—the idea that employers will replace humans with more efficient technology.

This anxiety is real. It is also largely invisible in the professional literature.

Museum directors and archive managers are experimenting with AI transcription, metadata generation, and curatorial recommendation systems. Staff are watching. They are wondering: am I being replaced? And this question is being asked not in conferences but in team meetings, performance reviews, and resignation letters.

The sector is publishing guidelines on responsible AI deployment. It is not publishing on responsible management of staff transition, reskilling, or the psychological toll of watching your expertise become automatable.

The Liability Question We Are Avoiding

When an AI system misidentifies an object, miscatalogues a record, or generates false metadata—who is liable? Is it the institution that deployed the model? The vendor who sold it? The people who trained the dataset? The curator who failed to catch the error?

No one is publishing comprehensive guidance on liability. Most institutions are moving forward with handshake agreements and hope.

The Cost-Benefit Analysis We Are Not Doing

What does it cost to adopt AI responsibly? Six months of metadata standardisation. Staff training. An external audit. Ongoing oversight and correction. Many small and mid-sized institutions cannot afford this. So they either do not adopt, or they adopt without preparation.

There is a growing gap between institutions that can afford responsible AI adoption and those that cannot. No one is discussing this affordability crisis in public forums.

❓ Big Question

If the curator's role is to exercise discernment, to author interpretation, to make subjective choices about meaning and value—but AI systems increasingly make those choices faster and at scale—how do you preserve curatorial authority in an automated world?

And more uncomfortable: if you cannot preserve it, should you be hiring curators?

💬 About

I'm Alfie, a researcher and archaeologist exploring where heritage, ethics, and AI meet. This digest keeps things short, critical, and useful—no jargon, no hype.

👉 Read or subscribe at glamorousai.beehiiv.com

👉 Send papers, reports, or ideas for August. And if your institution is wrestling with these questions, I'd like to hear your story (anonymously, if you prefer).