Back to MITRE ATLAS
MITRE ATLAS · AML.T0051
LLM prompt injection
“Adversaries craft instructions in inputs or retrieved content to alter an LLM's behavior or exfiltrate data.”
ATLAS technique AML.T0051
deployment
Cross-framework relationships
Lattice uses an explicit, four-type vocabulary and refuses ‘equivalent.’ Each relationship below carries a rationale and an editorial reviewer credit.
partially addresses
1- ManageNIST AI RMF · MANAGE
ATLAS techniques like prompt injection are managed (not eliminated) under NIST MANAGE; documenting residual adversarial risk is required.
Reviewer: Editorial: K. Park · Reviewed 2026-05-02
Version: Pinned 2026.Q1Last reviewed: 2026-05-02