Record summary
A quick snapshot of what this page covers.
Techniques22Attack methods connected to this risk.
Mitigations21Defenses that may help with related attacks.
Domain2. Privacy & SecurityThe broad risk area this belongs to.
Risk profile
How this risk is described and categorized.
Domain2. Privacy & Security
Subdomain2.2 > AI system security vulnerabilities and attacks
Entity1 - Human
Intent1 - Intentional
Timing2 - Post-deployment
CategoryMisuse tactics to compromise GenAI systems (Data integrity)
SubcategoryData exfiltration
Suggested mitigations
Defenses that may help with related attacks.
Limit Model Artifact Release
Business and Data UnderstandingDeployment
Control Access to AI Models and Data at Rest
Business and Data UnderstandingData Preparation+2 more
Encrypt Sensitive Information
Data PreparationML Model Engineering+1 more
AI Model Distribution Methods
Deployment
Code Signing
Deployment
AI Telemetry Logging
DeploymentMonitoring and Maintenance
Privileged AI Agent Permissions Configuration
Deployment
Single-User AI Agent Permissions Configuration
Deployment
AI Agent Tools Permissions Configuration
Deployment
Human In-the-Loop for AI Agent Actions
Deployment
Restrict AI Agent Tool Invocation on Untrusted Data
Deployment
Segmentation of AI Agent Components
DeploymentBusiness and Data Understanding
Input and Output Validation for AI Agent Components
Business and Data UnderstandingData Preparation+1 more
Restrict Number of AI Model Queries
Business and Data UnderstandingDeployment+1 more
Control Access to AI Models and Data in Production
DeploymentMonitoring and Maintenance
Sanitize Training Data
Business and Data UnderstandingData Preparation+1 more
Validate AI Model
ML Model EvaluationMonitoring and Maintenance
AI Bill of Materials
Business and Data UnderstandingData Preparation+1 more
Maintain AI Dataset Provenance
Data PreparationBusiness and Data Understanding
Verify AI Artifacts
Business and Data UnderstandingData Preparation+1 more
Generative AI Guardrails
ML Model EngineeringML Model Evaluation+1 more
Source
Research source for this risk, when available.
Included resource
Generative AI Misuse: A Taxonomy of Tactics and Insights from Real-World Data
Original source
MIT AI Risk Repository
Open the public repository used for AI risk records and taxonomy fields.