Shadow AI is AI usage that falls outside an organization's approval or management controls.
The term describes a relationship between an activity and the organization's rules. It is not a separate kind of model or a property permanently attached to an application. The relevant boundary can concern the feature, account, purpose, data, or permitted actions.
Known application, unapproved use
An approval defines the circumstances in which people or systems may use an application. Those circumstances matter because permission for one function does not automatically extend to another.
Hypothetical example: a company approves a document editor for collaborative drafting but explicitly excludes its optional AI summarization feature pending review. Using that feature falls outside the approval even though the editor itself is known and authorized. If the approval record instead says nothing about the feature and no applicable rule resolves its status, the reviewer has an unanswered question, not evidence of a deliberate violation.
What the label does not establish
Shadow AI does not, by itself, prove that someone acted maliciously, that information leaked, or that activity went unrecorded. Each conclusion requires evidence about the event in question. An approval gap and a confirmed security incident describe different findings, although they can occur together.
An unknown status also needs its own label. If the available evidence cannot establish which account, feature, or approval applied, uncertainty should remain explicit. Calling the activity approved would overstate the evidence; calling it prohibited could do the same.
How to clarify the approval boundary
Compare the actual use with a decision that specifies its scope. The useful question is whether that decision covers this feature and context, rather than whether the application's name appears on a list. Identify who can resolve an ambiguity or authorize a change; visibility alone cannot make that decision.
NIST's AI RMF Playbook, GOVERN 1.6 recommends mechanisms for inventorying AI systems and assigning responsibility for maintaining that inventory. An inventory supports this work, but an entry is not a substitute for a scoped approval.
Related concepts
- AI governance is the broader system of organizational policies, responsibilities, controls, and evidence within which approval decisions are made.
- AI usage governance concerns the conditions under which people and systems use AI. Shadow AI identifies activity that falls outside the applicable approval or management boundaries.
The Shadow AI pillar explains how to assess incomplete signals, establish the relevant context, and turn a finding into an accountable policy decision.