When AI is Built In: A Library Ethics Check-In

If you’ve updated any software in the last year — Microsoft 365, Google Workspace, Canva, Zoom, or Adobe to name a few — you’ve probably noticed something new. A search enhancement here, a meeting summary generator there, a design tool that now suggests layouts based on your organization’s past work.

These features are arriving inside the tools library professionals rely on every day, built in by vendors who are, in most cases, genuinely trying to be transparent about what they’re adding and why. The challenge isn’t bad intent. It’s that the technology is moving fast enough that none of us — libraries or vendors — have settled on a shared set of best practices for how embedded AI should work, how it should be labeled, what data it trains on, and how much control users should have over it.

As noted in the 2026 Library Systems Briefing, published in June 2026 in American Libraries Magazine, the library technology industry is at a “critical juncture” and AI is the main driver of that disruption. Major vendors have already embedded AI capabilities across their platforms, and the trend is accelerating across the board. Meanwhile, in July 2026, ALA Council adopted formal guidance for the use of AI in libraries, organized around the core values that have always guided the profession: public good, intellectual freedom, privacy, sustainability, DEIA, and labor.

As many libraries are still wrestling with whether it’s ethical to use AI, and where the guardrails are, they are likely already using it in ways they don’t even realize. But that’s okay. It’s not too late to take a step back and build some intention into our use of AI — from chatbots to content creation to built-in features in our everyday software.

Where AI Is Showing Up

It’s tempting to think of AI in libraries as a question about whether to use GhatGPT or not. And it is — but it’s also a question about the tools you use to do everything else.

Library-specific tools. ILS platforms, discovery layers, and chatbot interfaces are increasingly shipping with AI features: natural language search, automated metadata generation, recommendation engines, and patron-facing chat assistants. These are the most visible, and they’re where library vendors have made the heaviest AI investments.

Productivity and office tools. Microsoft 365 has Copilot embedded in Word, Excel, Teams, and Outlook. Google Workspace has Gemini in Docs, Gmail, and Drive. These tools process your library’s strategic plans, budget documents, personnel communications, and patron correspondence — often with default settings that may send data to external servers for processing.

Design and content tools. Canva, Adobe Creative Suite, and other design platforms now offer AI-powered layout suggestions, image generation, and content drafting. For the library staff creating flyers, social graphics, and programming materials, these features are convenient — but they raise questions about training data, attribution, and bias in generated imagery.

Communication platforms. Zoom’s AI Companion, Slack’s AI features, and Teams’ intelligent recap all summarize conversations, surface action items, and search across message history. These are powerful tools for busy library teams, but they also mean that internal discussions, patron inquiries, and collaborative decisions are being processed by AI systems.

The common thread: in every case, the AI arrived as part of an update, not as a separate purchase. And in many cases, the default settings are permissive by design.

Ethics for Embedded AI

When thinking about how to apply enthics to built-in AI tools, ALA guidance gives us a strong starting point. Here’s what each of its six values asks us to consider when evaluating the AI features already on our desktops.

Public Good

ALA guidance states that AI in libraries should advance equitable access to information and serve the public interest — not just vendor efficiency or institutional convenience. When evaluating an embedded AI feature, ask: does this tool help us serve our community better?

Intellectual Freedom

AI systems that curate, filter, or rank content can introduce bias — sometimes invisibly. The ALA guidance calls for transparency about how AI influences what patrons see and don’t see. For library-specific tools, this means understanding how search algorithms rank results. For productivity tools, it means being aware that AI-generated summaries or suggestions may be innacurate or incomplete.

Privacy

This is perhaps the most immediate concern. AI features in Microsoft 365, Google Workspace, Zoom, and Slack often process data on remote servers. The ALA guidance prioritizes tools that minimize data collection, offer opt-in rather than opt-out, and protect patron confidentiality. Before enabling any embedded AI feature, library staff should know:

• What data does this feature send to third-party servers?
• Is that data used for model training? (Related reading: Open vs. Closed: The Battle for the Future of Language Models)
• Can the feature be disabled without breaking core functionality?
• What happens to data processed by the AI after the interaction ends?

Sustainability

AI has significant environmental costs — training and running large models consumes substantial energy and water. The ALA guidance encourages libraries to favor tools with lower environmental impact and to ask vendors to disclose energy consumption.

Diversity, Equity, Inclusion, and Access

AI can perpetuate and amplify existing biases, from racial and gender bias in language models to accessibility gaps in AI-generated content. The ALA guidance calls for evaluating tools for fairness, accessibility, and representation. This is especially important for patron-facing AI features, where biased outputs can directly impact community members. But it also applies to internal tools, which should be evaluated with the same care.

Labor

The ALA guidance is clear: AI should augment rather than replace library workers. Embedded AI features should make professional work easier and more effective, not erode professional judgment or eliminate positions. Libraries should be transparent with staff about how AI affects workflows and job roles, and vendors should be asked how their tools are designed to support — not supplant — library professionals.

---

What to Ask Vendors (and Your Own Institution)

The ALA guidance includes practical recommendations for vendor conversations. Here’s a checklist adapted for embedded AI:

1. Can this feature be disabled without breaking core functionality? If not, that’s a design choice worth questioning.
2. What data is sent to third-party APIs, and is it used for training? Get this in writing. Vague privacy policies are not sufficient.
3. Is AI processing opt-in or opt-out? Opt-in respects patron and staff autonomy. Opt-out defaults should be a red flag.
4. How are AI-generated outputs labeled? Patrons and staff deserve to know when they’re interacting with AI-generated content, summaries, or recommendations.
5. What is the environmental impact of this feature? If the vendor can’t answer, note that and follow up.
6. How does this tool handle bias and accessibility? Ask for documentation on testing and mitigation.
7. What happens to our data when we stop using this product? Data portability and deletion matter, especially for AI-processed content.

A Shared Responsibility

None of this is about rejecting AI. Many of these features are genuinely useful — they save time, reduce repetitive work, and open up new possibilities for service. The point is that we need shared norms for how embedded AI should work, how it should be disclosed, and how much control libraries should have over it.

Vendors building in good faith want to know what libraries need. Libraries using these tools want clear standards to guide their decisions. The ALA guidance gives us the values framework. The check-in questions above give us a practical starting point.

The next step is conversation — between libraries and vendors, between library leadership and staff, and between libraries and the communities they serve. That’s what this post is meant to start.

If you’re a library professional navigating embedded AI in your own workplace, I’d love to hear what questions you’re asking and what answers you’re getting. And if you’re a vendor reading this — reach out. We’re all figuring this out together!