Risk
Insufficient Understanding of AI System Boundaries
Description
Failure to clearly define the complete boundaries of a discovered AI system, including all its components, interfaces, and direct dependencies.
Example
An AI-powered recommendation engine is identified, but its reliance on a separate, less secure microservice for data ingestion is missed.
Assets Affected
AI App
Model Inference endpoint
Pipeline Job
3rd-party AI integration
Mitigation
- For each AI system, meticulously map its architecture, components, and all internal/external interfaces
- Document system, computing resources, and tooling resources
Standards Mapping
- ISO 42001: A.6.2.3, A.4.2
- NIST AI RMF: MAP 2.1, MAP 4.1