Show HN: ModelFuzz – Open-source runtime guardrails for AI agents

Category: security

Tags: ai-security, llm-agents, prompt-injection, guardrails, red-teaming

Score: 6.3/10 (Innovation: 6, Technical: 6, Documentation: 5, Utility: 7)

ModelFuzz offers runtime guardrails for AI agents by combining an offensive scanner that red-teams OpenAI-compatible endpoints with a defensive decorator that enforces policies on tool arguments. It targets the emerging need for safety and security in LLM-powered agents, providing a practical toolkit for both vulnerability discovery and prevention.

Target audience: AI engineers, security researchers, backend developers building LLM agents

Repository: https://www.modelfuzz.com/

View on Hacker News