Humanbound

Humanbound is an open-source security testing platform that continuously evaluates AI agents and chatbots against adversarial attacks. Install via pip, point it at your agent's HTTP endpoint, and receive a security posture score (0-100, A-F graded) with findings mapped to OWASP LLM and Agentic Top 10 frameworks. Multi-turn adaptive testing covers prompt injection, jailbreaks, goal hijacking, tool misuse, data exfiltration, and PII disclosure. Integrate with your CI/CD pipeline using the native pytest plugin to enforce security gates and catch regressions. Use your choice of LLM provider—Azure OpenAI, OpenAI, Anthropic, Gemini, or custom. Pair with the open-source humanbound-firewall to block detected attacks in production. Apache 2.0 licensed, free for commercial use, with optional Community dashboard for posture tracking and compliance reporting (MITRE ATLAS, NIST AI RMF, EU AI Act, ISO 42001, GDPR).