Open-Source AI Red Team Tool 'Decepticon' Released
PurpleAILAB has released Decepticon, a new open-source autonomous multi-agent AI red team testing service. The tool is designed to conduct advanced hacking simulations and automate aspects of the red teaming process. It is intended for integration into home labs and for professional tool exploration.
- The tool is built using the LangChain and LangGraph frameworks, allowing it to leverage the growing AI agent ecosystem for scalable and innovative security testing. - PurpleAILAB has termed the methodology "Vibe Hacking," a paradigm where AI agents, rather than human testers, autonomously execute red team operations. - Decepticon is designed with a flexible multi-agent architecture, currently enabling a "Swarm" model for direct agent collaboration, with plans for a centralized "Supervisor" architecture. - It supports a variety of AI models for its operations, including cloud-based APIs from OpenAI and Anthropic, as well as locally run models through Ollama. - The project is currently experimental and in active development, with the creators encouraging community bug reports, feature requests, and pull requests to improve stability and functionality. - AI red teaming tools like Decepticon are designed to uncover vulnerabilities specific to machine learning systems, such as prompt injection, data poisoning, and model manipulation, which are not typically found in traditional penetration testing. - The open-source space for AI red teaming includes other notable tools like Microsoft's PyRIT and the open-source Garak scanner, each designed to find security weaknesses in large language models. - The lab behind the project, PurpleAILAB, is also developing Vigilo, a separate AI-powered tool focused on the offensive security of Web3 smart contracts.
