Resources

Association of College and Research Libraries · October 2025

The foundational professional framework. Defines four competency areas (Ethical Considerations, Knowledge & Understanding, Analysis & Evaluation, Use & Application) and five guiding mindsets. Sub-competency level detail makes it actionable.

American Library Association · 2026

ALA's values-based framework for library AI, built on six core values: Public Good, Intellectual Freedom, Privacy, Sustainability, Diversity/Equity/Inclusion/Access, and Labor. Directs libraries to treat AI output as drafts needing human review, prohibits entering patron data into unapproved tools, and provides a concrete vendor data-review checklist. The anchor for the ethics, privacy, discovery, and vendor-evaluation modules in this curriculum.

Association of Research Libraries · April 2024

Seven guiding principles including the essential 'No Human, No AI' - consequential decisions must not be delegated entirely to AI. Required reading for anyone developing library AI policy. Still the April 2024 edition with no revision issued; ARL's newer AI work appears below in the 2026 workshop report and quick poll.

Karen Estlund & Cynthia Hudson Vitale · ARL + CNI · CC BY 4.0 · March 2026

Scenario-planning report stress-testing library strategy against four divergent 2035 AI futures, with a year-later scan of what libraries have actually done. Its four near-term priorities: invest in the workforce now, leverage unique collections, lead on governance and ethics, and build partnerships. Names agentic AI governance, vendor-embedded AI features, and provenance tracking as the least developed areas in practice. Openly licensed and free to quote.

Leo S. Lo & University of Virginia Library · CC BY 4.0 · 2026

The one openly licensed framework in this list, and therefore the one any institution can adapt directly. Separates irreversible model training from retrieval-based access, which is the distinction that lets a library approve one request and decline another. Core rule: irreversible models do not get access unless item-level provenance and meaningful attribution can be demonstrated, and the institution retains contractually enforceable control to stop further use. Ships with an Adoption Kit (doi:10.18130/jbeg-a995), also CC BY 4.0.

University of Virginia Library (coordinating) · April 2026

A voluntary twelve-month commitment for libraries, archives and museums responding to AI training requests for unique cultural collections. Signatories classify each request, evaluate digitization and AI use as separate agreements, require provenance, and apply a presumption against approval for broad commercial training. Launched with twelve founding institutions; signatories share an anonymized ledger of the requests peers are receiving. Where a librarian points when a training request actually arrives.

Stony Brook University Libraries · December 2025

A published, versioned library AI policy - a worked example rather than a framework. Eight guiding principles led by 'no human, no AI', and, unusually, an explicit list of what the libraries do not use AI for: no tracking or evaluating individual patrons including search and borrowing history, no determining service eligibility or replacing professional expertise, no selling sensitive data, no training or fine-tuning on sensitive information. The negative list teaches more than the positive one, because a commitment about what a library will not do can actually be checked.

Library Copyright Alliance (ALA, ACRL, ARL) · October 2025

Short position statement asserting that existing U.S. copyright law, including fair use, is adequate to the copyright and generative AI intersection and that no new legislation is required. Its practical use is as the counterweight to vendor licence terms: when a publisher's contract restricts uses that copyright law would permit, this is the document that says so.

Leo S. Lo · CC BY-NC 4.0 · 2025 preprint

Five components - Technical Knowledge, Ethical Awareness, Critical Thinking, Practical Use, and Societal Impact - across four progressive levels, with a self-assessment rubric. An ARL workshop report (2026) pairs this with the ACRL competencies as a shared baseline. Worth reading against this site's own three-level structure.

Leo S. Lo · Journal of Academic Librarianship 52(1), 103186 · January 2026

Classify, Assess, Review, Enhance: a method for the reference interaction that begins with a patron holding an AI answer rather than a question. Built on an answer typography sorting AI answers into four kinds - factual, interpretive, constructive, and strategic - on the principle that each one fails differently. Subscription required; request through interlibrary loan if your institution lacks access.

Leo S. Lo · portal: Libraries and the Academy 26(3) · 2026

Six components for designing and governing an AI agent: Purpose and Audience, Authority, Structure and Style, Tools and Access, Accountability and Safeguards, and Signals and Review. The nearest thing the profession has to a published agent-governance framework. Subscription required, though the six components are documented openly in UT Austin Libraries' AI guide.

Association of College and Research Libraries · Revised September 2026

Revised in September 2026 to address AI directly. The same six frames now include AI-related practices, such as disclosing AI use, recognizing that generative AI is not inherently authoritative, and weighing the environmental cost of AI-generated information. A ready vocabulary for teaching critical evaluation of AI output.

Rick Dakan & Joseph Feller · Anthropic Academy · CC BY-NC-SA 4.0 · 2025

A free 12-lesson course built around the 4D Framework for human-AI collaboration: Delegation, Description, Discernment, and Diligence. Tool-agnostic and designed to remain useful as AI evolves - not a list of prompts that will be outdated next month. The three-mode model (Automation, Augmentation, Agency) and the product/process/performance layers from this course are referenced throughout this curriculum. Full course with exercises and certificate at Anthropic Academy; video-only access via the YouTube playlist.

Toronto Metropolitan University Libraries · November 2025

An academic library that credentialed the same 4D Framework this curriculum is built on. Two core workshops - Artificial Intelligence: Delegation & Description, and Artificial Intelligence: Discernment and Diligence - with a badge attached. Described as building a robust, tool-agnostic skill set rather than a step-by-step guide, and framed to help participants use generative AI thoughtfully or choose not to use it at all. Evidence that the 4D grounding is professional practice, not one author's preference.

Clarivate / Choice

The best continuous source for AI in libraries news and analysis. Weekly content. The Clarivate partnership with ACRL/Choice means strong institutional grounding.

Ontario Council of University Libraries + LibTech Insights · Launched August 2025

Bimonthly blog posts evaluating AI tools for library work. Practical, current, neutral tone. Individual installments cover research assistants, speech-to-text, coding tools, task management, and more.

ALA / RUSA

The professional interest group specifically for AI in reference and user services. Programming, discussions, and community for practitioners doing this work.

Clarivate · Annual - most recent: 2026

Annual global survey of library technology adoption, now in its third year. The authoritative benchmark data for AI adoption rates, training gaps, and institutional confidence. Key 2026 findings (n = 1,876): librarian confidence has stalled at 3.2 out of 5, unchanged from 2025; 33% of libraries remain at the exploration and evaluation stage and only 3% are in active implementation; 49% have no defined strategic position on AI; 30% report no institutional focus on AI literacy, and 50% of librarians learn AI on their own. Top concerns are now privacy and security (64%), misinformation (60%), and research and academic integrity (53%).

Association of Research Libraries · Fielded January 2026, published May 2026

Snapshot of where ARL member libraries stand: implementation is outpacing governance, only a minority report formal enforced generative AI policies, and staff attitudes vary within teams rather than splitting into camps. Asked what support they need, leaders named governance templates, readiness assessment tools, culture-change support, and peer examples - not more AI tools. Read as directional only: 39 completed responses from 125 member libraries, and ARL itself flags likely selection bias.

Montana State University Libraries and partners · IMLS-funded · CC BY 4.0 · 2025

Not a checklist. A structured thirty-minute reflection exercise: choose an AI scenario, identify the three values that most concern you, then draw a stakeholder card at random representing a perspective different from your own, identify that stakeholder's three most pressing values, and work through the prompts. The random-stakeholder step is what makes it usable as a staff development activity rather than an individual evaluation tool. Interactive web tool plus a printable PDF, and openly licensed.

Independent

Visualizes the citation network around a research paper. Not generative AI, but AI-enhanced discovery. Excellent for showing students how scholarship is connected - a strong instruction tool.

Ought Inc.

AI research assistant that searches Semantic Scholar (real papers) and returns AI-summarized results. Lower hallucination risk than general AI because it grounds responses in actual search results. Strong for systematic review workflows.

Consensus AI

Searches peer-reviewed papers and synthesizes empirical findings. Good for quick evidence checks: 'Does exercise improve sleep quality?' - returns real papers with synthesized findings. Free tier available.

Perplexity AI

Combines AI generation with web search and citations. Every answer includes sources. Better for factual research queries than ChatGPT. Academic mode available. Lower hallucination risk on factual questions. Free tier functional.

Anthropic

Strong for long documents, nuanced writing, and instruction design. Handles large context windows well - paste in a full policy document and ask questions. More conservative tone by default. Claude for Work includes stronger data privacy protections.

OpenAI

Most widely used. The default model is capable for most library tasks. Largest user base means the most tutorials, community examples, and third-party guidance available. Free tier rate-limited; paid tier ($20/mo) gives more access to the slower, more capable Thinking mode.

Google

Best integration with Google Workspace - Docs, Drive, Gmail. Real advantage if your library runs on Google. Access to Google Search reduces factual errors on current-events queries. Review your institutional Google agreement for privacy terms.

Lovable AI

AI app builder for non-programmers. Describe what you want in plain language; Lovable builds a working web app. Good starting point for library tool-building. Used in Module 16.

Replit

Development environment with strong AI assistance. Good for iterating on small tools with AI help. Free tier available. Widely recommended in library AI literacy contexts (HKU LibGuide).

Zapier

No-code workflow automation connecting 9,000+ apps. Now includes AI actions and integrations. For library automation: connecting forms to spreadsheets, triggering emails, routing data between systems. Free tier for simple automations.

Make

Visual workflow automation platform, more powerful than Zapier for complex workflows. Better for multi-step automations. Has strong AI integrations. Free tier available for low-volume workflows.