AI Governance Engineer
GitHub
On this page

Pattern: Shadow-AI Discovery

Continuous discovery of AI systems and agents running without a registry entry, reconciled against the registry so the inventory matches production.

Layer 02 · Inventory & Transparency In the chapter 05 catalogue

One working control on the engineering side of AI governance.

Summary: Continuously discover AI systems and agents that are running but not registered, and reconcile them against the registry, so the inventory reflects reality rather than only what teams remembered to declare. You cannot govern what you cannot see.

Shadow-AI Discovery A workflow diagram generated by Archify. 01 / Where AI hides 02 / Layer 02 Inventory & Transparency 03 / Registry EX / Frozen 05 / Evidence Scan Match + triage Register + evidence Cloud accounts · running services · Where AI hides › Scan Cloud accounts running services Network egress · external API calls · Where AI hides › Scan Network egress external API calls Identity providers · agent logins · Where AI hides › Scan Identity providers agent logins Code repos · SDK, prompt use · Where AI hides › Scan Code repos SDK, prompt use Discovery scan · finds models, agents · Layer 02 Inventory & Transparency › Scan Discovery scan finds models, agents Match vs registry · known / unknown · Layer 02 Inventory & Transparency › Match + triage Match vs registry known / unknown Triage & claim · assign an owner · Layer 02 Inventory & Transparency › Match + triage Triage & claim assign an owner Registry entry · owner, scope · Registry › Register + evidence Registry entry owner, scope Scope frozen · until claimed · Frozen › Register + evidence Scope frozen until claimed Discovery report · reconciliation · Evidence › Register + evidence Discovery report reconciliation unknown as evidence reconcile unclaimed: freeze open entry, claim Legend Agent logic Policy Context / trace External system
Shadow-AI DiscoveryDiscovery scans the places AI hides, matches each finding against the registry, and turns unknowns into registry entries or blocks, leaving a discovery report. Run it before you claim your inventory is complete. Generated from the Body of Knowledge.Open interactive diagram (opens in a new tab)

Objectives

Close the gap between the registry and production by finding unregistered models, agents and AI-enabled tools, and bringing them under governance.

Target users

AI governance engineer, security engineer, platform team.

Impacted stakeholders

Model owners, security operations, auditors.

Relevant principles

Register and bound every actor before it acts; make the governed path the easiest path.

Context

An organisation where teams adopt AI tools and spin up agents faster than any central inventory can track, and where, by one vendor’s comparison of the category, much of the AI-governance platform market “manages the program … without any runtime data path”1.

Problem

A registry fed only by voluntary declaration is always behind. Unregistered agents, the shadow fleet, are exactly the layer a paper inventory cannot see, and a security vendor’s 2026 survey reports that roughly one in eight AI breaches involved agentic systems2.

Solution

Run discovery against the environments where AI appears (identity providers, cloud accounts, network egress, code repositories, SaaS integrations) using discovery tooling (illustrative) to find models and agents. Reconcile findings against the registry, open an entry for each unknown with an owner to claim it, and escalate the unclaimed. Feed the result back into the Agent Registry’s drift check.

Consequences

The inventory converges on reality and the blind spot shrinks. The cost is discovery integration and the process to triage and claim what it finds.

Agent Registry; Agent Identity & Scoped Credentials; Continuous Assurance Telemetry.

Maps to: EU AI Act Art. 49/71 · ISO/IEC 42001 · NIST AI RMF (Map) · CSA AICM · OWASP Agentic ASI10 · Layer 02 Inventory & Transparency.

Threat IDs follow the OWASP Top 10 for Agentic Applications 2026 3 and function labels the NIST AI RMF4. Mappings are illustrative, not a claim of conformity.

Sources

  1. [1] “Best AI Governance Platforms in 2026: 14 Enterprise Vendors Compared” (vendor-published comparison of the 13 Magic Quadrant vendors plus its own product; most of the category “manages the program … without any runtime data path”). Kosmoy. 2026-07-10. https://www.kosmoy.com/resources/blog/best-ai-governance-platforms-2026/ (verified: secondary)
  2. [2] 2026 AI Threat Landscape Report (vendor survey; key finding stated on the report page: one in eight breaches were agentic). HiddenLayer. 2026. https://www.hiddenlayer.com/report-and-guide/threatreport2026 (verified: primary)
  3. [3] Top 10 for Agentic Applications 2026 (ASI IDs). OWASP GenAI Security Project. 2025-12-09. https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/ (verified: primary)
  4. [4] AI Risk Management Framework (AI RMF 1.0; Govern, Map, Measure, Manage). NIST. 2023-01-26. https://www.nist.gov/itl/ai-risk-management-framework (verified: primary)
Edit this page on GitHub
Cite this pattern

García Aibar, J. (2026). Pattern: Shadow-AI Discovery. In AI Governance Engineering: The Thesis & Body of Knowledge (v0.5.0), chapter 05, Patterns. https://doi.org/10.5281/zenodo.22956197. https://aigovernanceengineer.com/patterns/shadow-ai-discovery. CC BY 4.0

BibTeX

@misc{aige2026bok,
  author  = {Jorge García Aibar},
  title   = {{AI Governance Engineering: The Thesis \& Body of Knowledge}},
  chapter = {05. Patterns: Shadow-AI Discovery},
  year    = {2026},
  version = {0.5.0},
  doi     = {10.5281/zenodo.22956197},
  url     = {https://aigovernanceengineer.com/patterns/shadow-ai-discovery},
  note    = {Version 0.5.0}
}
Share on LinkedIn

Obligations this pattern evidences

Register rows that list this pattern among the patterns that build their evidence. Mappings are illustrative, not a claim of conformity.

Cases where it applies

Incident post-mortems that name this pattern as a control that would have caught the failure.

Draft controls in the open control profiles that build on this pattern. A control states the requirement and the evidence it must leave behind.