The Archives Are Getting AI-Ready
Slowly. Too Slowly
The good news is that the archival community has moved beyond vague talks about AI. Frameworks are taking shape, and task forces are not just proposed but are actively working. Funding is starting to flow in ways that suggest institutions understand this is not a fleeting technical interest but a fundamental shift.
The less good news is that the pace still feels like it’s set by someone leisurely crossing a green light while traffic builds in all directions. There is progress, but not with urgency, and definitely not the kind that matches the moment.
So, it’s worth pausing briefly to see what has actually changed and what has not.
The profession finally appears to be organizing — at least on paper.
The Society of American Archivists quietly noted in its February Council meeting minutes that it reviewed a draft charge for a new group focused on AI and archives, with a special call for task force appointments coming in March for the 2026–2027 term. That’s nothing new. SAA moves cautiously, and establishing a standing group is a meaningful sign of where the profession is putting its institutional weight.
Meanwhile, the latest issue of American Archivist features research evaluating thirty-three software tools to see how AI can enhance accessibility and discoverability in special collections. Having thirty-three tools reviewed in a peer-reviewed journal is practical, grounded work that the field really needs — not just think pieces about AI’s potential, but honest appraisals of what works in actual collections.
The FLAME guidelines are solid. Now, someone needs actually to use them.
A recent notable development is the release of the Archives and Records Association AI Preparedness Guidelines, which signal a shift from general concern about AI toward more practical, operational thinking. These guidelines respond to mounting pressure to implement AI in archival settings without adequate grounding in archival principles, data quality, or ethics, aiming to connect data readiness, archival theory, and responsible AI within a workable framework.
That diagnosis is accurate, and the guidelines are a valuable contribution. But this is where the familiar gap reappears. A framework isn’t an intervention unless institutions have the staffing, funding, and mandate to implement it. Most do not. The backlog problem isn’t solved by describing it more clearly or aligning it more neatly with AI readiness. It’s solved by increasing capacity, and that remains the real obstacle.
“Cold data” is gaining attention now — and archives should pay heed.
One overlooked aspect: the enterprise storage world is waking up to something archivists have known forever — that old data holds value. Even long-dormant archival “cold” data might contain valuable insights when fed into an AI model. The commercial sector is suddenly very interested in making historical data machine-accessible, which creates both opportunities and pressures for cultural heritage institutions. The opportunity: more tools, more vendor interest, and more infrastructure investment. The pressure: if enterprises figure out how to unlock their cold data before archives do, archives risk seeming like a solved problem that nobody prioritized in time.
The Internet Archive situation continues to worsen.
This situation keeps escalating. Major publishers are now actively blocking the Internet Archive crawlers over copyright disputes tied to AI training data, and this action is already causing real damage to preservation. Archived pages often provide the only reliable record of how stories originally appeared. Publishers routinely edit, change, or remove articles, which leaves the Internet Archive as the sole source for tracking those changes.
What’s infuriating about this is that organizations like the Internet Archive are not building commercial AI systems. They’re nonprofits dedicated to preservation. They’re being penalized for being useful to researchers who share the same infrastructure as those publishers who are angry. While publishers have the right to defend their content, fighting that battle by dismantling the historical record is wrong.
Every director should remember this number:
Gartner predicts that 60% of AI projects will fail by 2026 due to a lack of AI-ready data. Most institutions are rushing into AI experiments without doing the metadata work, identifying gaps, or laying the necessary groundwork, and they will have a lot to explain to their boards in about eighteen months. The archivists who advise, “get your collections in order first,” will be proven right in the most frustrating way.
The uncomfortable truth nobody wants to voice out loud:
The main problem with AI readiness in archives isn’t mainly about technology. It’s about labor and funding, disguised as a tech issue. The FLAME guidelines are excellent, the SAA task force is a good sign, and research published in the American Archivist is exactly what’s needed. But many organizations still lack formal document management processes. Fixing that requires investing in the people who do the work. Until funders, administrators, and policymakers connect AI readiness with the ongoing underfunding of description and processing work, we’ll keep producing guidelines that remain ineffectual.
More on this next time. Let me know what I’m missing — especially if you’re working on any of this from inside an institution.