ziran

ZIRAN is a comprehensive security testing framework designed to identify vulnerabilities in AI agents, including those with complex capabilities such as tool usage and memory. By modeling agents as graphs of capabilities, it effectively discovers dangerous tool chains, detects execution-level side effects, and conducts adaptive multi-phase campaigns, surpassing the capabilities of single-prompt scanners. Notable features include graph-based analysis, tool-chain discovery, and thorough coverage of established security benchmarks like OWASP and MITRE.