Skip to main content

MITRE ATLAS

MITRE ATLAS (Adversarial Threat Landscape for AI Systems) is a globally-accessible knowledge base of adversary tactics and techniques based on real-world observations of attacks on machine learning systems. VirtueRed comprehensively tests AI systems against 14 tactical categories that represent the complete adversarial attack chain targeting ML/AI infrastructure.

Overview

ATLAS provides a comprehensive model for understanding how adversaries compromise and operate within AI environments, enabling organizations to assess their AI systems' resilience against the full spectrum of ML-specific threats—from initial reconnaissance through data exfiltration and impact. The framework examines how AI systems might be exploited at each stage of an attack, recognizing that AI introduces unique vulnerabilities through model endpoints, training pipelines, data repositories, and integration points.

Tactical CategoryDescription
ReconnaissanceGathering information about target ML systems
Resource DevelopmentAcquiring resources and developing attack tools
Initial AccessGaining initial foothold in ML systems
ExecutionRunning adversarial code or actions
PersistenceMaintaining access across restarts and updates
Privilege EscalationGaining elevated access or capabilities
Defense EvasionAvoiding detection and bypassing security
Credential AccessObtaining credentials and authentication tokens
DiscoveryExploring the AI environment
Lateral MovementMoving through connected systems
CollectionGathering data from target systems
Command and ControlMaintaining communication with compromised systems
ExfiltrationExtracting data from ML systems
ImpactDisrupting, degrading, or destroying ML systems

Reconnaissance

Adversary efforts to gather information for targeting AI systems and planning attacks.

TechniqueDescription
Active ScanningTesting exposure of AI endpoints to network reconnaissance
Search Open Technical DatabasesEvaluating information leakage in public repositories
Phishing for InformationAssessing susceptibility to targeted information gathering
Search Open Websites/DomainsTesting exposure of AI documentation and specifications
Gather Victim Identity InformationEvaluating protection of AI team member information

Testing Approach:

  • System prompt extraction attempts
  • API boundary probing
  • Model capability inference testing
  • Access control enumeration

Resource Development

Adversary efforts to establish resources for operations against AI systems.

TechniqueDescription
Develop CapabilitiesTesting detection of adversarial ML tool development
Obtain CapabilitiesAssessing awareness of publicly available AI exploitation tools
Acquire InfrastructureEvaluating detection of attack infrastructure setup
Compromise InfrastructureTesting identification of supply chain targeting
Establish AccountsAssessing detection of fake accounts for AI service access

Testing Approach:

  • Jailbreak technique evaluation
  • Adversarial input generation
  • Attack vector assessment

Initial Access

Techniques to gain initial foothold in systems hosting AI components.

TechniqueDescription
Exploit Public-Facing ApplicationTesting vulnerabilities in AI API endpoints and interfaces
Supply Chain CompromiseAssessing risks from compromised ML libraries and frameworks
Valid AccountsEvaluating use of legitimate credentials to access AI services
PhishingTesting targeted attacks against AI teams and administrators
Trusted RelationshipAssessing exploitation of third-party integrations and partnerships

Testing Approach:

  • Direct prompt injection testing
  • Input sanitization evaluation
  • Authentication boundary testing
  • Dependency security assessment

Execution

Techniques for running malicious code in AI environments.

TechniqueDescription
Command and Scripting InterpreterTesting code injection through AI notebooks and shells
User ExecutionAssessing social engineering targeting AI developers and operators
Container Administration CommandEvaluating exploitation of containerized AI deployments
Serverless ExecutionTesting abuse of serverless AI functions and lambdas
Cloud Administration CommandAssessing execution through cloud management interfaces

Testing Approach:

  • Code execution boundary testing
  • Action authorization evaluation
  • Plugin security assessment

Persistence

Techniques for maintaining access to AI systems across restarts and updates.

TechniqueDescription
Account ManipulationTesting persistence through compromised AI service accounts
Create AccountAssessing unauthorized account creation in AI platforms
Scheduled Task/JobEvaluating persistence through training pipeline schedules
Server Software ComponentTesting backdoors in model serving infrastructure
Container Orchestration JobAssessing persistence in Kubernetes-based AI deployments

Testing Approach:

  • Behavioral consistency testing
  • Context carryover evaluation
  • Session persistence assessment
  • Model backdoor detection

Privilege Escalation

Techniques for gaining higher-level permissions in AI systems.

TechniqueDescription
Valid AccountsTesting escalation through compromised administrator accounts
Cloud Service DashboardAssessing privilege escalation via cloud console access
Container API ExploitationEvaluating escalation in containerized AI environments
Access Token ManipulationTesting privilege elevation through token modification
Domain Policy ModificationAssessing manipulation of AI governance and access policies

Testing Approach:

  • Role-based access control testing
  • Permission boundary evaluation
  • Privilege escalation attempts

Defense Evasion

Techniques to avoid detection and bypass AI security measures.

TechniqueDescription
Obfuscated Files or InformationTesting evasion through encoded payloads and scripts
Indicator RemovalAssessing deletion of attack evidence from AI audit logs
Impair DefensesEvaluating disabling of AI security monitoring and alerts
MasqueradingTesting impersonation of legitimate AI processes and services
Valid AccountsAssessing use of compromised legitimate credentials to avoid detection

Testing Approach:

  • Filter bypass evaluation
  • Obfuscation technique testing
  • Encoding variation assessment
  • Incremental attack simulation

Credential Access

Techniques for stealing credentials used for AI system access and authentication.

TechniqueDescription
Unsecured CredentialsTesting exposure of AI API keys and service tokens
Credentials from Password StoresAssessing theft from credential management systems
Steal Application Access TokenEvaluating extraction of OAuth tokens and JWT credentials
Steal Web Session CookieTesting session hijacking for AI management platforms
Brute ForceAssessing password attacks against AI interfaces and services

Testing Approach:

  • Credential leakage testing
  • Token exposure evaluation
  • Session security assessment

Discovery

Techniques for exploring the AI environment to map systems and identify valuable targets.

TechniqueDescription
System Information DiscoveryTesting enumeration of AI infrastructure and configurations
File and Directory DiscoveryAssessing discovery of models, datasets, and code repositories
Network Service DiscoveryEvaluating identification of AI endpoints and APIs
Cloud Service DiscoveryTesting enumeration of cloud-based AI services
Permission Groups DiscoveryAssessing mapping of AI platform roles and privileges

Testing Approach:

  • Model fingerprinting attempts
  • Capability enumeration
  • Integration boundary testing

Lateral Movement

Techniques to move through the AI environment and connected systems.

TechniqueDescription
Remote ServicesTesting movement through SSH/RDP to AI infrastructure
Cloud Service DashboardAssessing lateral movement via cloud management consoles
Software Deployment ToolsEvaluating abuse of CI/CD pipelines for AI models
Taint Shared ContentTesting propagation through shared models and datasets
Use Alternate Authentication MaterialAssessing movement using stolen tokens and keys

Testing Approach:

  • Cross-system access testing
  • Model pipeline exploitation
  • Service integration boundary assessment

Collection

Techniques for gathering information of interest from AI systems prior to exfiltration.

TechniqueDescription
Data from Information RepositoriesTesting extraction of training datasets and model artifacts
Data from Cloud StorageAssessing theft of cloud-hosted models and datasets
Screen CaptureEvaluating capture of AI dashboard and monitoring interfaces
Automated CollectionTesting systematic extraction of AI assets and configurations
Archive Collected DataAssessing preparation and staging of stolen AI resources

Testing Approach:

  • Training data extraction attempts
  • PII leakage testing
  • System prompt recovery evaluation
  • Historical data access assessment

Command and Control

Techniques for communicating with and controlling compromised AI systems.

TechniqueDescription
Application Layer ProtocolTesting covert channels through AI API endpoints
Data ObfuscationAssessing hidden communications in legitimate AI traffic
Dynamic ResolutionEvaluating use of dynamic infrastructure for C2
Web ServiceTesting command channels through legitimate cloud services
Encrypted ChannelAssessing use of encryption to hide malicious communications

Testing Approach:

  • Covert channel detection
  • Instruction injection evaluation
  • State manipulation assessment

Exfiltration

Techniques for stealing models, data, and intellectual property from AI systems.

TechniqueDescription
Exfiltration Over C2 ChannelTesting data theft through established command channels
Exfiltration Over Web ServiceAssessing use of legitimate services for data transfer
Automated ExfiltrationEvaluating systematic model and dataset extraction
Transfer Data to Cloud AccountTesting exfiltration to attacker-controlled storage
Exfiltration Over Physical MediumAssessing theft via removable media or local access

Testing Approach:

  • Model extraction resistance testing
  • Data leakage prevention evaluation
  • Knowledge boundary assessment

Impact

Techniques to disrupt, corrupt, or destroy AI systems and operations.

TechniqueDescription
Data DestructionTesting deletion of critical models and training datasets
Data Encrypted for ImpactAssessing ransomware attacks targeting AI assets
Service StopEvaluating disruption of AI inference and training services
Resource HijackingTesting cryptomining or compute theft using AI infrastructure
Data ManipulationAssessing integrity attacks on models and training data

Testing Approach:

  • Behavior hijacking evaluation
  • Resource exhaustion testing
  • Harmful content generation assessment
  • Reputation risk evaluation

See Also