NAAMSE

NAAMSE is an automated security fuzzing framework designed to evaluate vulnerabilities in LLM-based agents using evolutionary algorithms. By generating adversarial prompts through intelligent mutations, it tests for security issues such as jailbreaks and prompt injections, employing a behavioral scoring engine and organizing attack vectors through a clustering engine. Key features include comprehensive reporting of vulnerabilities and metrics, making it an essential tool for red-teaming and enhancing the security posture of LLM agents before deployment.