Humanbound

Humanbound is an AI agent security testing platform that can be run as a Python command line tool or through a hosted platform. A user points it at the endpoint of a chatbot or agent, defines a scope or lets the tool extract one, and receives findings mapped to OWASP LLM and agentic AI categories. Tests cover prompt injection, jailbreaks, tool abuse and data exfiltration in single turn, multi turn and agentic modes, and an LLM as a judge evaluation scores the results against the defined scope. It is aimed at security and engineering teams that build customer support bots, internal knowledge assistants and other generative AI systems. The platform reports a security posture score from 0 to 100 with an A to F grade, tracks vulnerabilities through open, stale, fixed and regressed states, and exports guardrail rules to OpenAI or a Humanbound firewall format. A pytest plugin and exit codes allow insecure deployments to be blocked in continuous integration pipelines. The project describes itself as open source and offers a local engine that needs no login. Pricing: The vendor does not publish plan prices in its documentation.