About
Humanbound is an open-source security testing platform for AI agents and chatbots. It runs continuous adversarial tests against live endpoints, judges every response, and maps findings to OWASP LLM and Agentic Top 10 standards. Install via pip, point it at any agent, and get an A-F security posture score. Features include multi-turn adaptive attacks covering prompt injection, jailbreaks, goal hijacking, and data exfiltration. Native pytest plugin adds AI security gates to CI/CD with baseline comparison. Bring your own LLM from OpenAI, Azure, Anthropic, Gemini, or custom providers. Findings integrate with the open-source humanbound-firewall for runtime protection. Apache 2.0 licensed and free for commercial use. Free Community plan offers posture history, findings tracking, and hosted dashboard. Built for AI, security, and platform teams in regulated industries. Findings also map to MITRE ATLAS, NIST AI RMF, EU AI Act, ISO 42001, and GDPR for compliance evidence.