Defining Shadow EdTech
Shadow EdTech describes AI tools operating inside course delivery platforms and learning management systems without formal IT or security approval. This includes AI chat assistants invoked during live instruction, browser-based writing aids used by students, summarization extensions installed by instructors, and third-party plugins connected to an LMS through open integration standards.
The term extends the established concept of shadow IT, where unsanctioned software enters an organization outside procurement and security review, to the specific context of course and training environments. What distinguishes shadow EdTech from earlier shadow IT problems is agent-like behavior: many of these tools do not simply store or display data; they read course content, generate responses, and transmit information to external AI vendors, a function that traditional application allow-listing was not designed to evaluate.
How Unapproved AI Tools Enter Course Environments
Unapproved AI tools typically reach course environments through four technical paths. Each path is designed for legitimate extensibility. The risk is not the integration mechanism itself, but the absence of a review step that evaluates what an AI tool does with the data it can reach.
Browser extensions
AI writing aids and chat tools with page content or clipboard access, installed directly by instructors or students.
LMS plugins and LTI connectors
Third-party tools integrated through open marketplace standards and Learning Tools Interoperability style integrations.
API and OAuth connectors
Standing access granted to instructor or admin accounts, which can persist beyond the original use case.
Embedded widgets
External tools added directly into course pages via iframe without platform-level registration.
Browser extensions request permissions such as page content access or clipboard access, letting them read course material or student submissions inside the browser session. LMS plugin marketplaces and LTI integrations allow external tools to exchange data with a course platform through a supported but often lightly reviewed integration point. Embedded iframe widgets let an instructor add a third-party tool directly into a course page. API connectors, authorized through OAuth grants tied to an instructor or administrator account, give an external AI service standing access to course or LMS data.
Where Shadow EdTech Hides in the Stack
Unapproved AI activity does not sit in one neat log stream. It cuts across session, identity, network, and application layers, which is why single-control detection often misses it.
-
Session layer
Browser extension-based AI tools operate at the session layer, outside LMS-native logging, so their activity may not appear in platform audit logs even when data is exposed.
-
Identity layer
OAuth grants issued to third-party AI plugins are tied to a user's identity and can remain active after a course ends or an account is deprovisioned, creating long-lived unmanaged access.
-
Network layer
Outbound calls to AI vendor domains can be visible at the perimeter or proxy layer, but encrypted traffic and shared CDN infrastructure make reliable identification difficult.
-
Application layer
LMS plugin marketplaces and LTI connectors operate within the platform, but many institutions lack a centralized inventory of which integrations are active and what data they can access.
Detecting Unsanctioned AI Tool Activity
Detecting shadow EdTech usually depends on correlating signals across three layers rather than relying on any single control.
- Network traffic analysis can surface outbound connections to known AI service endpoints, though this signal alone produces false positives against benign productivity tools and false negatives against encrypted or proxied traffic.
- Identity and access monitoring, specifically new OAuth application authorizations against an LMS or course-platform identity provider, can surface an unapproved AI tool connection at the moment it is granted, before it is used at scale.
- Endpoint telemetry, including browser extension inventories, can be compared against an approved software list to flag AI tools installed outside a managed process.
None of these signals is sufficient on its own. A tool discovered on the network but with no corresponding identity grant may be benign; a token that persists after an account is deprovisioned represents access without corresponding usage. Reliable detection requires combining discovery, data-access classification, and enforcement rather than treating any one signal as conclusive.
Data at risk. Course platforms often process student records, assessment data, and employee training records. When an unapproved AI tool has read or transmit access to this data, exposure can occur without appearing in LMS-native audit logs, since much of this activity happens at the browser session or API layer.
Where Runtime Governance Applies
Traditional application approval processes, built for installable software, generally stop at a binary allow or block decision made before a tool is used. AI tools operating inside course environments behave differently: they act on data continuously, which means governance has to extend past the approval step into the tool's ongoing behavior.
Runtime governance addresses this gap by evaluating what an AI tool does with course or LMS data as it operates, rather than only at install or authorization time. This includes establishing an identity for each AI agent or tool, scoping its permissions to least privilege, enforcing policy on the actions it is permitted to take, and maintaining audit logs of its activity. Trussed AI provides this layer of runtime governance and security for enterprise AI agents, applying agent identity, permissions, and policy enforcement to AI tools operating within enterprise systems, including course and training platforms where unsanctioned AI activity has been identified.