Model Extraction Attacks

Test AI systems for model theft vulnerabilities where attackers can reconstruct models through queries.

Quick Reference

Skill:       model-extraction
Agent:       04-llm-vulnerability-analyst
OWASP:       LLM03 (Supply Chain), LLM02 (Sensitive Info Disclosure)
MITRE:       AML.T0024 (Model Stealing)
Risk Level:  HIGH

Extraction Techniques

1. Query-Based Extraction