Deploy AI agents you can prove are secure
Humanbound is an open-source security layer for AI agents. It enables developers and security teams to run automated adversarial tests before launch, block prompt injections and policy violations at runtime via a firewall, and continuously monitor posture scores mapped to OWASP, NIST AI RMF, and EU AI Act frameworks.
- Audience
- Security leaders and AI engineers building and deploying production LLM agents who need provable compliance and security.
- Category
- ai-security, developer-tools, open-source
- Key features
- Automated adversarial & behavioral pre-launch testing across 180+ scenariosSingle-turn, multi-turn, and agentic attack mode simulationsHumanbound Firewall runtime protection to block prompt injections before reaching modelsLiving 0–100 posture scores with regression and drift detectionFramework mapping to OWASP LLM Top 10, OWASP Agentic AI, NIST AI RMF, and EU AI ActBlack-box API testing requiring no SDK or agent modificationsAir-gapped and local self-hosting support using OllamaExportable evidence packs in HTML, PDF, JSON, SARIF, and CEF formatsSIEM-native CEF events and webhook integrations for issue tracking