As artificial intelligence tools become pervasive, public libraries may want to establish transparent guidelines for how they are used by staff

Policy statements are important, because “people have very different ideas about what is acceptable or appropriate,” says Nick Tanzi, assistant director at South Huntington Public Library (SHPL), NY, who was recently selected by the Public Library Association to be part of a Transformative Technology Task Force focused on artificial intelligence (AI).

In the library field, opinions about AI—particularly with the recent emergence of large language models (LLMs) such as ChatGPT, Gemini, Claude, and Copilot—currently run the gamut from enthusiastic adoption to informed objection. But even the technology’s detractors would agree that AI has already become an integral part of the information-seeking tools many people use every day. Google searches now frequently generate Gemini AI responses as top results. Microsoft has ingrained Copilot into its Windows OS and Office software. ChatGPT’s global monthly active users exceeded 800 million at the end of 2025. Patrons are using these tools, and they may have questions or need assistance. Libraries should be clear about how these and other AI technologies are being used within their institutions.

Tanzi notes that this may just require a review. “Do I need a new policy, or can I see in this case an emerging technology through an existing policy lens?” he says. Can library policies on issues such as patron privacy or data security simply be adapted to address AI? “We’re always looking at existing policy with an acknowledgement of this new elephant in the room.”

While policies tend to be written by individuals or small groups of administrative staff, Tanzi notes that AI is “a disruptive technology that impacts everybody differently, and because of that, having multiple perspectives can really help us address it more completely.” SHPL has a staff AI user group that was originally organized to demystify the technology, facilitate conversations, answer questions, learn about staff concerns, and discover how staff might want to use it. The group has also helped inform AI policy. “You never want a policy that is too disconnected from the frontline staff,” Tanzi says. “That can happen when you’re in administration. You tend to be less of the practitioner. A broad perspective from a lot of staff in a broad array of positions can really help.”

FOR YOUR CONSIDERATION

LLM AI technology and the ways staff and patrons encounter and use it have been evolving rapidly, so libraries may want to consider framing any AI policy as a living document that will require regular audits and updates—possibly as often as every six months—and ensure that the policy is flexible enough to apply to emerging AI-driven technologies. The Holderness Free Library, NH, for example, states in its AI policy that “AI applications and systems will be regularly evaluated” and that the “policy will be reviewed in a timely fashion to adapt to new developments.”

Establishing approved and prohibited uses of AI for library work can help set parameters for staff and clarify an institution’s use of the technology for patrons. (Some policies, such as Toronto Public Library’s (TPL) AI policy, specify that “individual customer use of AI tools is out of [its] scope.”) Low-risk workflow uses such as brainstorming program ideas, summarizing reports, correcting grammar in documents, or drafting emails could be listed as acceptable, while a library may want to state within its policy explicit prohibitions against using AI for creating “deepfakes” or other content that might be used to mislead the public, for patron surveillance, or for hiring decisions and employee performance reviews. The St. Charles Public Library, IL, includes a specific prohibition against using “AI alone to make employment decisions. This includes decisions regarding the hiring, discipline, suspension, termination, promotion, demotion, or salary increase/decrease of employees. AI should not be used to evaluate or conduct performance reviews.”

The DeKalb Public Library, IL, outlines acceptable and prohibited uses with a policy establishing that AI can be used “to assist patrons in retrieving information from library catalogs, databases, and digital collections, improving search accuracy and efficiency”; to provide user assistance by answering “common questions and directing patrons to appropriate resources”; to “personalize content recommendations for patrons based on their preferences”; to “assist with the creation of content for informational or marketing purposes” (with the stipulation that this content must be reviewed by staff for accuracy and compliance with the library’s brand and marketing guidelines); and for analyzing “library usage data, such as circulation statistics and patron demographics, to inform decision-making and improve services.” Prohibited uses of AI in DeKalb’s policy include any use of AI “for surveillance purposes within the library premises, including facial recognition or tracking patron behavior without their explicit consent,” discriminatory practices, and any unauthorized data collection.

More broadly, AI policies should conform with a library’s existing privacy and security policies. This may require explicit prohibitions against entering any personally identifiable patron information such as names, addresses, or library card numbers into AI tools, since many free, publicly available AI tools store inputs for use as training data. But at minimum, use of patron information with AI should be transparent, and patron consent should be obtained. The staff policy for AI use at Houston County Public Library System, GA, describes AI models as unauthorized third parties with which library employees should never share a patron’s protected personal information or personally identifiable information “as such models are not capable of keeping any data shared with them private,” and states that doing so would be a violation of the library’s existing personnel policy, Georgia confidentiality law, and Georgia libraries’ Public Information Network for Electronic Services (PINES) policy.

Also, if a library plans to deploy an AI tool for patrons, or if it will be regularly used by staff, it should be vetted for security practices. The U.S. Department of Commerce’s National Institute of Standards and Technology’s AI Risk Management Framework, published in 2023, is one freely available standard, and is cited by TPL’s policy. Other considerations might involve encouraging staff to opt out of AI training features when possible, and checking document metadata for staff or patron names, geolocation, and other identifiable information before uploading files to an AI tool.

In addition to a list of appropriate uses for AI similar to DeKalb’s, the Hastings Public Library, MI, has a staff AI usage policy that emphasizes that it is important to double-check a document’s metadata before uploading it into an AI tool: “When uploading documents to be analyzed or summarized, ensure the documents do not contain any personally identifiable information or Library data.... When asking an AI to analyze any datasets, ensure the data has been anonymized,” it reads. “Documents should be checked not only for content, but for data about the document (metadata), such as who created it.”

Austin Public Library has formed an AI Implementation Team

IN THE LOOP

Given the current tendency of LLM AIs to sometimes hallucinate/generate incorrect responses or omit important information when summarizing documents, libraries may want to define “human-in-the-loop” accountability, such as requiring mandatory fact checking or cross-referencing of AI-generated content or responses used by staff. After explaining AI hallucinations in its policy, the Kenosha Public Library (KPL), WI, states that “due to the potential for realistic and convincing hallucinations, critical evaluation of all GenAI outputs is essential.”

At minimum, libraries should clearly disclose when AI is being used to assist with patron questions or generate other information given to patrons—including the use of AI to help create library marketing materials, newsletters, or other public-facing content such as blog posts and other library website content. And staff should ensure that AI outputs don’t violate any existing library policies.

KPL also specifies in its AI policy that attribution statements are not required when AI is used for tasks such as brainstorming or grammar and spell-checking. However, “when AI is used to assist in the creation of original works such as blogs, social media posts, reports, summaries, or marketing materials, acknowledgement of the role AI plays in the creation process is required. All original work in which AI plays a role in its creation will link to a statement...indicating the use of AI.” The library also currently prohibits the use of AI text-to-image tools for the creation of patron-facing content.

In addition, there are emerging concerns about the copyright and human creativity implications of AI, many of which are explored in “The Copyright Conversation” in this month’s issue (pp. 31–33). Tanzi, however, suggests that public libraries may want to avoid attempting to establish any blanket prohibitions against AI-generated content in their collections at this stage. “Collection development policy is one area that I think may have to be fine-tuned or reimagined, but you don’t want to go and say, ‘We will absolutely 100 percent not allow AI generated materials.’ It’s difficult to detect...so absolutism doesn’t necessarily work with respect to that. Policy tends to be more nuanced—what is enforceable and what is the intent behind it.”

Libraries that are already offering classes and programs to help patrons understand and navigate AI tools and AI-generated content may want to consider including ongoing staff and/or public education efforts as a component of any new policy.

REINVENTING THE WHEEL

A growing number of public libraries of all sizes have published AI policies that can help guide libraries planning to write one of their own. However, since some states and municipalities are beginning to pass AI regulations, finding examples within a library’s home state will probably be the most helpful.

In Texas, Austin’s city council had begun coordinating the creation of local policies and the deployment of AI solutions at city departments and agencies, and Austin Public Library (APL) was closely involved with those efforts. However, last June, Texas Gov. Greg Abbott signed the Texas Responsible AI Governance Act (TRAIGA) into law. “High-risk” AI systems—prohibiting the development and deployment of AI for purposes such as behavioral manipulation, social scoring, discrimination, the creation or distribution of deepfakes or child pornography, or any use that would infringe on constitutional rights—are one point of focus for the law. But the law also supersedes any city legislation regulating AI within the state.

“Our understanding is that we can suggest best practices for city use. But we cannot create policy which limits how these tools can be used, outside of the ones which the state says are unallowed,” explains Will Harlan, Librarian III Innovate Supervisor for APL, who was the library’s representative on the city council’s AI project and is currently chairing the AI Implementation Team for APL. Aspects of the policies the council was considering, such as risk management processes, requirements for human review and oversight, and restrictions on data sharing, can become guidelines at APL and in Austin city agencies, but unless regulations are included in TRAIGA, they won’t be legally enforceable.

With the help of its AI Implementation Team, APL still plans to “reinforce AI readiness and literacy for the whole library, prepare front-facing staff on how to help our customers and our patrons with questions about AI, establish sensitive ethical guidelines for AI adoption and use, and export opportunities for AI use within each of the library’s divisions,” Harlan says. Like Tanzi, he notes that AI could have “really far-reaching implications in terms of nearly every aspect of how we operate,” so APL has included staff from multiple departments in the process.

Before forming the team, Harlan also sent a questionnaire to staff to get the pulse of their opinions on AI. He discovered that “it’s a pretty big mix. There are a lot of people who are super excited, there are a lot of people who are incredibly concerned, and there are also a lot of people who are both.... [They] realize that there are risks involved in this, and it’s worth at least talking about those so that we [can implement] it safely and reasonably, rather than either A) burying our heads in the sand or B) doing something like insisting or requiring our staff to use AI. There’s plenty of sky in between those extreme responses.”

Public libraries are well-suited to offer patrons assistance with new technology and connect other local government agencies with expert advice. But whether a library wants to establish a new set of policies or encourage the incorporation of AI into workflows, “I’d certainly urge people to slow down,” Harlan says. “The companies who have been first adopters when it comes to implementing AI haven’t gotten to the promised land of productivity [gains].... There’s no need to rush into this blindly.”