AI Resilience Report forArchivists

If you're worried that AI will replace archivists, here's some reassuring news: the field is largely treating AI as a helper, not a replacement. The U.S. National Archives (NARA) recently published an inventory showing several real, live AI projects — like using Azure OpenAI to automatically generate tags and topics for approximately 2 million digital records, which frees up staff to focus on other priorities, and a pilot that uses AI to auto-fill descriptive metadata and tackle "the descriptive gap" of labor-intensive cataloging ^[1]. NARA is also piloting AI tools that screen records for personally identifiable information before public release ^[1].

On the access side, OCLC researchers report that institutions are experimenting with AI for tasks like captions and transcriptions, relying on the strengths of large language models, though most teams still want AI built into the workflow tools they already use rather than as a standalone product ^[2]. Handwritten-text recognition platforms like Transkribus are widely used to decode old documents, and Penn State just received part of an $11 million Schmidt Sciences grant to build humanities-driven AI tools that expand and support Black digital archives ^[3]. The clear pattern: AI is automating the slow, repetitive parts (tagging, transcription, PII detection), while archivists keep the judgment-heavy work — appraisal, ethics, outreach, and managing collections.

Adoption is happening, but carefully. The UK's Archives & Records Association recently warned in its AI Preparedness Guidelines that "AI can support archival work, but only when collections are made 'AI-ready'" and that automation is "a constrained necessity, not a magic solution" ^[4]. Several factors slow things down: many collections have inconsistent metadata, copyright and privacy rules are complex, and LLMs can hallucinate facts about history — risks the Association of Canadian Archivists highlights when it calls for a "radical empathy" approach with attention to consent, power, inclusivity, and transparency in AI governance ^[5].

Cost is another brake; OCLC found that using an LLM for full entity reconciliation "would have been prohibitively expensive" ^[2] at Yale. On the other hand, the profession is organizing itself fast — the Society of American Archivists just launched an AI Task Force charged with helping the archival community "navigate the ethical, technical, accessibility, and organizational impacts of AI technologies" ^[6] and is drafting core AI competencies for the profession. Job demand remains steady: the Bureau of Labor Statistics projects employment of archivists, curators, and museum workers to grow 6 percent from 2024 to 2034, faster than the average for all occupations ^[7].

So the human skills — ethical judgment, community trust, and storytelling through outreach — are exactly what AI can't replicate, and they're what the field is doubling down on.