Library Use of Artificial Intelligence

Background
This policy establishes the standards that staff and representatives of The Seattle Public Library shall observe when acquiring and using Artificial Intelligence (AI) technologies to perform Library functions and deliver Library services. These standards are intended to ensure effective, secure, and responsible practices that are consistent with the Library’s mission, principles, and values.

Definitions
Artificial Intelligence (AI): A machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. AI systems use machine and human-based inputs to: perceive real and virtual environments; abstract such perceptions into models through analysis in an automated manner; and use model inference to formulate options for information or action.

Generative AI: A class of AI systems, including algorithms, deep-learning, and machine learning models, capable of generating new content (text, images, video, and audio) based on the inputs of training datasets. This also includes systems capable of ingesting input and translating that input into another form, such as text-to-code systems.

Policy Statement
AI technologies have the potential to solve meaningful challenges and enhance the delivery of Library services. However, they also have the ability to make mistakes, falsify information, and rely on biased or faulty data. Due to their design, the operations of AI systems are more difficult to predict, inspect, decode, or understand than traditional technology systems. Failures in, or inappropriate uses of, AI systems can erode public trust and cause harm to people and communities, potentially at a greater scale than with traditional technologies. All AI technologies acquired, configured, developed, operated, or maintained by or on behalf of the Library shall align with requirements outlined in this policy.

1. Use and Operation

• Library staff shall acquire and use only those AI products that have been approved for Library use via a process developed by the AI Governance Committee. Approval is required for free software, software-as-a-service (SaaS) products, web-based software services, and AI components that are embedded within or added to non-AI software products, even if those products have otherwise been approved.
• In determining if an AI product is approved for Library use, the AI Governance Committee shall:
       o Determine if the product is consistent with all relevant IT and Library standards and policies.
       o Document technical information about the AI models used in the product by requiring a vendor to provide a completed GovAI AI Factsheet, by consulting an existing AI Factsheet on file with the GovAI AI Registry, or by using an equivalent process.
• Library staff or its representatives shall not input Library data that has been classified as “sensitive” or “confidential” into an AI product or service without approval by the AI Governance Committee or via an exception process developed by the Committee.
       o If the classification of data is absent or unclear, the employee shall assume it is sensitive and seek further clarification and approval from the AI Governance Committee or the authorized steward of the data in question.

    B. Oversight and Accountability

• The Library shall establish an AI Governance Committee, which will be tasked with reviewing requests to acquire and use AI technologies, assessing how the practical use of these approved tools will be in keeping within the guardrails defined by policy, and making regular recommendations on how the organization should adjust its practices in this space.
• Outputs of Generative AI systems must be reviewed by a Library employee who is familiar with the subject matter prior to use in an official Library capacity, including through patron reference question responses, online publication, or other sharing methods with the public, partners, and others.
• Output reviews shall determine that the output is accurate, is free from errors or “AI hallucinations,” does not contain unlicensed intellectual property, is not discriminatory or biased against protected classes, and meets Library standards for communication.
• Output reviews must be documented and shall demonstrate how the review was conducted, how the determination to use the output was made, and who made the determination.
• Documentation of output reviews shall be retained by the employee who conducted the review, and must be produced upon request to any Library employee seeking relevant information in an official capacity.
• The Library may establish a central repository for output review documentation.

    C. Attribution

• All material produced by Generative AI systems that is published by or on behalf of the Library must be attributed to the AI product that was used, including the specific model and its version. Attributions must also denote that the material was reviewed by a Library employee.
       o Example attribution: “This blog post was co-written with OpenAI's GPT-3.5 model and was reviewed for accuracy by a member of Library staff.”
       o Attribution is not required for material that was initially produced by a Library employee or an authorized party on behalf of the Library, and was merely altered or edited by AI for spelling, grammar, usage, or style unless those modifications are substantial or significantly alter the nature of the original material.
• Computer code generated using AI by a Library employee and used in software, databases, scripts, formulas, or other automation functions must be documented and attributed via comments in the source code or, if necessary, in separate product documentation. In such documentation, the name of the Library employee who is responsible for the AI-generated code must be included.
       o Example attribution: “Code was generated using Anthropic Claude Code 1.0.32 and was reviewed for accuracy by Mike Wallace.”

     D. Records Retention and Disclosure

The Library is subject to RCW 40.14, which governs Preservation and Destruction of Public Records for state and local agencies, and RCW 42.56, known as the Washington State Public Records Act, which requires agencies to make public records available to members of the public with limited, narrow exemptions. Library staff must follow the laws and current procedures and policies for records retention and disclosure.
Retention requirements are based on the content of the records, not their format. Regardless of the required retention period, if records responsive to a public disclosure request exist, they must be disclosed.
When using an AI system, Library staff must preserve or destroy records created when using AI products pursuant to the relevant retention schedule, and are responsible for searching for and retrieving them if a public disclosure request is received.

     E. Prohibited Uses

The following uses of AI technology are prohibited.

• Identity Determination or Behavioral Prediction. Applications that use computer vision or other AI/machine-learning techniques to analyze or classify people’s facial and body characteristics or movements into emotions or sentiments, or that attempt to determine an identifiable person’s protected class information, or predict an identifiable person’s future behaviors.
• Misleading or Deceptive Content. Creation or distribution of any content that was created with the purpose or intent of misleading or deceiving Library employees or members of the public.

HISTORY
Adopted: July 31, 2025