MITRE ATLAS

Overview

ATLAS (Adversarial Threat Landscape for Artificial Intelligence Systems) is MITRE's knowledge base of adversarial tactics and techniques for machine learning systems. Think of it as ATT&CK but specifically for AI/ML.

URL: https://atlas.mitre.org

Tactics (High-Level Objectives)

TacticObjectiveTraditional ATT&CK Equivalent
ReconnaissanceGather information about the ML systemReconnaissance
Resource DevelopmentAcquire resources for the attack (compute, data, models)Resource Development
ML Model AccessGain access to the target modelInitial Access
ExecutionRun adversarial techniques against the modelExecution
PersistenceMaintain access or influence over the ML systemPersistence
EvasionAvoid detection by ML-based defensesDefense Evasion
ImpactDisrupt, degrade, or destroy ML system integrityImpact
ExfiltrationExtract information from the ML systemExfiltration

Key Techniques

Technique IDNameDescription
AML.T0000ML Model Inference API AccessInteracting with the model's prediction API
AML.T0004ML Artifact CollectionGathering model artifacts (weights, configs, code)
AML.T0010ML Supply Chain CompromisePoisoning models, data, or tools in the supply chain
AML.T0015Evade ML ModelCrafting inputs to evade ML-based detection
AML.T0016Obtain CapabilitiesAcquiring adversarial ML tools and techniques
AML.T0020Poison Training DataCorrupting the model's training dataset
AML.T0024Exfiltration via ML Inference APIExtracting data through model queries
AML.T0025Exfiltration via Cyber MeansStealing model artifacts through traditional methods
AML.T0040ML Model Inference API AccessUsing the API for extraction or evasion
AML.T0043Craft Adversarial DataCreating inputs designed to fool the model
AML.T0047ML-Enabled Product/Service AbuseAbusing AI features for unintended purposes
AML.T0051LLM Prompt InjectionInjecting adversarial instructions into prompts
AML.T0054LLM JailbreakBypassing model safety controls

Using ATLAS for Red Team Engagements

ATLAS maps directly to engagement phases:

  1. Scoping: Use ATLAS tactics to define test categories
  2. Planning: Map specific techniques to your target's attack surface
  3. Execution: Reference technique IDs in your testing notes
  4. Reporting: Cite ATLAS IDs in findings for standardized communication

Case Studies

ATLAS maintains a library of real-world incidents at atlas.mitre.org/studies. Review these for attack inspiration and to understand how techniques chain together in practice.