Chaos Agent

- **Chaos Agent** is an open-source AI agent framework for red-teaming LLM-based systems — it autonomously generates adversarial prompts to find vulnerabilities like jailbreaks, prompt injections, and data leaks before real attackers do. - **Red-teaming** means deliberately stress-testing an AI model to make it behave badly (bypass safety filters, reveal system prompts) so weaknesses can be fixed. - Unlike simple scripts, this is an **autonomous agent** that plans multi-step attacks and uses tools (web search, code execution) to probe deeper than a single query can. - It belongs to a growing family of automated AI security tools (alongside Garak and Microsoft's PyRIT) as companies rush to deploy LLMs in production without fully understanding their attack surface. - The same tool can be used defensively by researchers or offensively by malicious actors — a standard tension in open-source security tooling.