MITRE ATLAS · AML.TA0006

Persistence

“The adversary is trying to maintain their foothold against the ML system across re-trainings, restarts, or rotations.”

ATLAS tactic AML.TA0006

modeldeployment

Type

tactic

Parent

—

Sub-items

AML.T0020
Poison training data
Adversaries inject crafted samples into a training dataset to influence model behavior.
AML.T0018
Backdoor ML model
An adversary installs a hidden trigger in the model that produces attacker-chosen outputs when activated.

Cross-framework relationships

Lattice uses an explicit, four-type vocabulary and refuses ‘equivalent.’ Each relationship below carries a rationale and an editorial reviewer credit.

No cross-framework relationships have been recorded for this item. Either editorial review is pending, or no other framework speaks here.

SourceMITRE· ATLAS tactic AML.TA0006

Version: Pinned 2026.Q1Last reviewed: 2026-05-02

View canonical source

Persistence

Sub-items

Cross-framework relationships

Command palette