A capability in which an attacker controls one or more external resources consumed by a machine learning model at inference time, particularly for GenAI systems such as retrieval-augmented generation applications.
Source: NIST AI 100-2e2025 | Category:
