Record summary
A quick snapshot of what this page covers.
Techniques2Attack methods connected to this risk.
Mitigations6Defenses that may help with related attacks.
Domain2. Privacy & SecurityThe broad risk area this belongs to.
Risk profile
How this risk is described and categorized.
"If the input instructions themselves refer to inappropriate or unreasonable topics, the model will follow these instructions and produce unsafe content. For instance, if a language model is requested to generate poems with the theme “Hail Hitler”, the model may produce lyrics containing fanaticism, racism, etc. In this situation, the output of the model could be controversial and have a possible negative impact on society."
Domain2. Privacy & Security
Subdomain2.2 > AI system security vulnerabilities and attacks
Entity1 - Human
Intent1 - Intentional
Timing2 - Post-deployment
CategoryInstruction Attacks
SubcategoryUnsafe Instruction Topic
Suggested mitigations
Defenses that may help with related attacks.
Restrict Library Loading
Deployment
Code Signing
Deployment
Verify AI Artifacts
Business and Data UnderstandingData Preparation+1 more
Vulnerability Scanning
ML Model EngineeringData Preparation
User Training
Business and Data UnderstandingData Preparation+4 more
AI Bill of Materials
Business and Data UnderstandingData Preparation+1 more
Source
Research source for this risk, when available.
Included resource
Safety Assessment of Chinese Large Language Models
Original source
MIT AI Risk Repository
Open the public repository used for AI risk records and taxonomy fields.