Record summary
A quick snapshot of what this page covers.
Techniques26Attack 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
Timing1 - Pre-deployment
CategoryModel Attacks
SubcategoryPoisoning Attacks
Suggested mitigations
Defenses that may help with related attacks.
Memory Hardening
ML Model EngineeringDeployment+1 more
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
Code Signing
Deployment
Maintain AI Dataset Provenance
Data PreparationBusiness and Data Understanding
Model Hardening
Data PreparationML Model Engineering
Use Ensemble Methods
ML Model Engineering
Input Restoration
Data PreparationML Model Evaluation+2 more
Adversarial Input Detection
Data PreparationML Model Engineering+3 more
Verify AI Artifacts
Business and Data UnderstandingData Preparation+1 more
AI Bill of Materials
Business and Data UnderstandingData Preparation+1 more
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
Limit Model Artifact Release
Business and Data UnderstandingDeployment
Source
Research source for this risk, when available.
Included resource
Risk Taxonomy, Mitigation, and Assessment Benchmarks of Large Language Model Systems
Original source
MIT AI Risk Repository
Open the public repository used for AI risk records and taxonomy fields.