The Signal Digital Happenings at the Library of Congress
ISSN 2691-672X
- Home
- AI Sandbox Series: A Secure Environment to Explore Artificial Intelligence at the Library
August 31, 2026, Posted by: Leah Weinryb-Grohsgal
This post is the first in a series about the Library of Congress Artificial Intelligence (AI) Sandbox for staff. It provides an introduction to the initiative’s goals and an overview of the types of experiments conducted so far. Future posts will feature details about experiments in text recognition and searchability, metadata and bibliographic record enhancement, and more.
Citation: Highsmith, Carol M., 1946-, photographer Great Sand Dunes National Park & Preserve, one of America’s newest national parks to be established (in 2004), in the San Luis Valley at the base of the Sangre de Cristo Range. The dunes, the tallest in North America, were formed from sand and soil deposits of the Rio Grande River and its tributaries, flowing through the valley. Library of Congress Prints and Photographs Division Washington, D.C. 20540 USA http://hdl.loc.gov/loc.pnp/pp.print
As the steward of over 200 petabytes of data including manuscripts, maps, books, images, artwork, audiovisual recordings, web archives, legislative data, research reports, and copyright records, the Library of Congress has spent decades building infrastructure and services to preserve and provide access to digital materials. The predictive and generative capabilities of artificial intelligence technologies offer enticing possibilities for maintaining, delivering, and analyzing some of the nation’s most influential research and treasured cultural and historical artifacts. Yet the use of these technologies requires extensive evaluation and monitoring before implementation.
In 2025, the Library introduced its AI Sandbox for staff initiative to do just that. “Sandbox” environments have been used since the early 1970s to provide protected spaces for development and experimentation without risk to systems. Playing on the term for a secure and contained play space, as technology evolved rapidly in the 20th century, the sandbox setting was adopted to mean “an environment that is controlled and supervised to test new products and services.”
While AI technologies have growing potential to augment and enrich the Library’s collections and services, experimentation and security are key. To better understand how these creative and innovative methods might work with various materials and conditions, the most recent cohort of the Library’s AI Sandbox for staff supported over a dozen small-scale experiments. Teams incorporated technical and subject matter expertise to focus their experimentation on the real-life demands and possibilities for Library workflows.
Staff expertise remains core to the Library’s approach to AI. That expertise guided the formulation of these experiments, allowing staff to determine how to deal with any hurdles that arose and assess how human insight could lead the adoption of AI technologies.
Projects focused on materials spanning the Library’s rich collections: legal records and briefs; circulars and regulations; images; talking books; audiovisual materials, classified ads from historic newspapers; web archived government PDFs; manuscript materials in Arabic, Persian, Ge’ez, Turkish, Armenian, Georgian, Japanese, Chinese, and Hebrew; children’s and young adult books, sound recordings; web archived metadata; record annotations; oral history recordings; non-Latin monographs; catalog records; training materials and resources; and even more items accessible via the Library of Congress collections API.
The scale of these experiments varied, but all exceeded the information processing capabilities of even the most energetic human: millions of bibliographic records, hundreds of thousands of lines of data, tens of thousands of pages, thousands of minutes of video, thousands of books, and the vast loc.gov API materials. Teams explored AI capabilities and developed questions about its use.
Their goals included:
- Managing and speed up digitization processes, including identification, inventory, digitization, processing, and public presentation;
- Identifying and extracting key pieces of metadata and full text;
- Automatically generating summaries and topic lists;
- Reading handwritten text making it searchable and computationally analyzable;
- Transcribing materials and assigning subject terms;
- Creating research tools;
- Describing images to provide deeper access.
The AI Sandbox for staff initiative allows the Library to take imaginative ideas and evaluate them step-by-step, including their accuracy, scalability, costs, technical and administrative barriers, and risks. A testing ground for learning more about AI, this effort has enabled the Library to begin to see the range of ways staff envision using AI technologies to support our mission to engage, inspire, and inform Congress and the American people.
Stay tuned for future AI Sandbox posts, where we will explore efforts to evaluate AI used in text recognition and searchability, metadata and bibliographic record enhancement, transcript and document generation, and bibliographic access and research.
Categories
Continue/Read Original Article: AI Sandbox Series: A Secure Environment to Explore Artificial Intelligence at the Library | The Signal
Discover more from DrWeb's Domain
Subscribe to get the latest posts sent to your email.
