Record summary
A quick snapshot of what this page covers.
Techniques18Attack methods connected to this risk.
Mitigations8Defenses 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
Intent2 - Unintentional
Timing2 - Post-deployment
CategoryUnhelpful Uses
SubcategorySoftware Vulnerabilities
Suggested mitigations
Defenses that may help with related attacks.
Use Ensemble Methods
ML Model Engineering
Code Signing
Deployment
Memory Hardening
ML Model EngineeringDeployment+1 more
AI Bill of Materials
Business and Data UnderstandingData Preparation+1 more
Control Access to AI Models and Data at Rest
Business and Data UnderstandingData Preparation+2 more
Validate AI Model
ML Model EvaluationMonitoring and Maintenance
Privileged AI Agent Permissions Configuration
Deployment
Single-User AI Agent Permissions Configuration
Deployment
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.