Shadow AI is the use of AI tools or model call paths outside an organization’s approved governance. It can expose data and bypass security controls, even when an unapproved service produces the same outputs as an approved one. section 4.3.6
Shadow AI can enter through everyday tools or production systems:
- Employees use personal AI accounts, browser extensions, or mobile apps and paste in code, contracts, or customer data. section 4.3.6
- AI features appear in existing SaaS tools, or teams adopt libraries with embedded models without adding them to the organization’s AI inventory. section 7.4.3
- An engineer routes production requests directly to a model from another account, bypassing an approved gateway and its controls. section 4.3.6
Reduce the gap between sanctioned and actual use: train employees on policy, monitor usage through enterprise browsers or proxies, and provide approved alternatives that meet users’ needs. For compliance, inventory known AI systems and maintain operational evidence such as logs and tested oversight. section 4.3.6 section 7.4.3

