2026 EDUCAUSE Top 10
#5: Knowledge Management for Safer AI

Mitigating the risks of artificial intelligence by integrating knowledge management into data governance, privacy, and ethics programs

min read


Knowledge management can help institutional teams become more familiar with their enterprise artificial intelligence tools while also restricting those tools' access where needed. Knowledge Management for Safer AI is issue #5 in the 2026 EDUCAUSE Top 10.

Credit: Zach Peil / EDUCAUSE © 2025

Back to main article

When enterprise AI solutions fail, as they often seem to, poor data quality is likely a top contributing factor.Footnote1 AI solutions are only as good as the data they are trained on, and institutions that use incomplete, outdated, or overly complicated data are likely to see disappointing results that weaken campus stakeholder trust in those solutions.

To be most effective, our enterprise AI tools need to "know" us. They must understand, in data form, an institution's unique historical and cultural contexts, its particular business processes and policies, and its array of services, resources, and people. Enterprise AI tools that know these things about an institution can more effectively respond to prompts and arrive at solutions that are accurate and fitted to their institutional environments.

Enterprise AI tools that know us too well, on the other hand, introduce uncomfortable risks most institutions don't (and shouldn't) want. (We all know what happens in 2001: A Space Odyssey.) There are systems and types of data that institutions don't want their AI tools to access and share with others, so privacy and security concerns are paramount.

Knowledge management (KM) can help institutional teams become more familiar with their enterprise AI tools while also restricting those tools' access where needed. KM processes can help an institution organize and document its collective identity, structures, and practices, as well as articulate its values and goals for certain areas of AI inquiry and support. This, in turn, can lead to AI solutions that are more accurate and grounded in the institution's particular context. These KM processes are deeply collaborative and require engagement with stakeholders across the institution who can provide input on, and help evaluate, AI knowledge related to institutional systems, practices, and policies.

As campus-wide engagement with these tools grows, institutions will need to continuously align their KM processes with broader data governance and ethics efforts. This alignment will help to establish appropriate boundaries for AI tools while building consensus on the types of information various stakeholder groups will require from those tools. It will also clarify the types of information stakeholders don't need to—or should not—access.  

Campus Spotlight: Collaborative Chatbot Development at UW–Madison

Didier Contis, vice provost for Information Technology and chief information officer at the University of Wisconsin-Madison (UW-Madison), said the adoption of AI technologies across the campus has introduced challenges in ensuring these tools consistently and accurately provide guidance and support to students and other users. "As we see more and more people across our campuses adopting or trying out AI technology," he asked, "how do we start to have the discussion about ensuring consistency, particularly when the tool is supposed to help guide a student through a financial aid application or navigate a program's curriculum?"

The solution, what Contis calls "knowledge curation," is a process of engaging with and gathering input from the institution's vast network of practitioners and knowledge experts, each of whom can help shape and validate the institution's AI outputs. As UW-Madison continues to explore its AI capabilities, new processes are being developed to engage with this network earlier on to help ensure chatbot outputs are accurate and tailored to the institution's unique context.

"In some of our pilot efforts to develop an AI chatbot to guide employees on policies—such as sick leave or research grant application—we feed the AI a massive amount of policy documentation from our policy library," Contis said. "While these pilots have provided valuable insights into making policy content more user-friendly, a fundamental issue persists: Some essential contextual information is missing from the policy documents. And this is where you start to realize the limitations. This has shown us the need to engage knowledge experts in specific policy domains to test and validate the quality of the answers."

Ways to Get Started

Through our panelist interviews and community survey, technology leaders noted some ways institutions might establish knowledge management for AI:

  • Form an AI governance body or committee to continually evaluate the accuracy and safety of the institution's AI tools and outputs. In particular, these governing bodies should consider documenting "sensitivity levels" across the institution's data, establishing categories of sensitivity that will help determine which types of data are appropriate for AI use and which are not. The University of Notre Dame's Information Security Policy, for example, includes "security designations" for the institution's data.
  • Knowledge management of enterprise AI tools will require clear and updated documentation of institutional policies, practices, services, and resources that can serve as data inputs and the foundation of what our AI tools "know." Many institutions will discover that their existing documentation (e.g., policies, web materials) is outdated, incomplete, or insufficient for reliable AI outputs. Taking stock of the institution's knowledge base and identifying these and other gaps may be a necessary first step for sound KM.

Note

  1. Enterprise Data Leaders: Our AI Ambitions Are Stalling Out on Silos, (Reltio, December 2024). Jump back to footnote 1 in the text.

Didier Contis is Vice Provost for Information Technology and Chief Information Officer at University of Wisconsin-Madison.

Isabel Gallin is Vice President of EUNIS and Relationship Manager at Karlsruhe Institute for Technology, Germany.

Justin Gatewood is Chief Information Security Officer at California Community Colleges Technology Center.

Hemalatha Manickavinayaham is Associate Vice President of Planning and Digital Transformation at California State University, Sacramento.

John McGuthry is Vice President and Chief Information Officer at California State Polytechnic University, Pomona.

Brandon Rich is Director of AI Enablement at University of Notre Dame.

© 2025 EDUCAUSE and the 2025–2026 EDUCAUSE Top 10 Panel. The text of this work is licensed under a Creative Commons BY-NC-ND 4.0 International License.