Show HN: Busted – eBPF tool that monitors what your AI agents send (github.com)

Busted is a newly unveiled eBPF-based tool designed for monitoring and managing communications between large language models (LLMs) and various AI service providers, such as OpenAI and Anthropic. Developed in Rust, this high-performance observability system operates at the kernel level, allowing for real-time tracking of AI agent behavior without necessitating any changes to the applications themselves. A standout feature is its ability to enforce policies, such as blocking specific content sent to APIs, exemplified by a command that restricts certain keywords from being transmitted to Anthropic's API. This tool is significant for the AI/ML community as it provides a robust means of governance in increasingly complex AI environments while ensuring compliance and security. Technical capabilities include TLS plaintext capture, behavioral classification using machine learning, and integration with security information and event management (SIEM) systems. Busted supports Kubernetes, enhancing its utility in containerized environments. By minimizing the performance overhead and operating transparently, it opens the door for enterprises to maintain a rigorous check on AI communications, promoting responsible usage and enhancing data security—a crucial aspect as AI continues to evolve and integrate into diverse applications.