Record summary
A quick snapshot of what this page covers.
Techniques13Attack methods connected to this risk.
Mitigations13Defenses 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 (Model integrity)
SubcategorySteganography
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
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
Code Signing
Deployment
Restrict Library Loading
Deployment
Vulnerability Scanning
ML Model EngineeringData Preparation
User Training
Business and Data UnderstandingData Preparation+4 more
Verify AI Artifacts
Business and Data UnderstandingData Preparation+1 more
AI Telemetry Logging
DeploymentMonitoring and Maintenance
Input and Output Validation for AI Agent Components
Business and Data UnderstandingData Preparation+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.