We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1
LLM Security
Quick Overview Security Recommendations
Vulnerabilities Manipulate LLM via special Least privilege, Zero trust, Prompt Injection input Input/Output validation LLM outputs without Insecure Output Zero trust, Proper output proper validation/ Handling validation and sanitization sanitization ML-BOM attestation, Data Training Data Polluted data sources sanitization, Zero trust, Poisoning compromises integrity MLSecOps Make LLM unavailable to Input validation, Rate Model Denial of Service use limiting, Monitoring Tampered training data, Audit Data Sources, Supply Chain vulnerable s/w Vulnerability Management, Vulnerabilities components Code Signing Input Validation, Output Sensitive Information LLM outputs sensitive, Redaction, Restrict external Disclosure proprietary information data sources LLM plugin accepts inputs Input Validation and Insecure Plugin Design from the model without Parameterization, Adhere validation API best practices Least Privilege, Avoid open- Excessive Functionality, Excessive Agency ended functions, Permissions, Autonomy Downstream Authorization LLM provides erroneous Oversight, Continuous Overreliance information Validation, Risk Disclaimers Robust Access Controls, Unauthorized access of Model Theft Centralized Model Inventory, LLM Zero-trust, Watermark