uGen: From Gaps to Exploits — A Multi-Agent, Retrieval-Augmented System for End-to-End Microarchitectural Attack Code Generation

uGen is a multi-agent, retrieval-augmented generation (RAG) system designed to automate the end-to-end lifecycle of microarchitectural attack research. Given a high-level security objective, uGen autonomously identifies unexplored attack surfaces, retrieves relevant academic and technical knowledge, generates functional attack code, validates it against real hardware, and iteratively improves both the attack and potential defenses. This work advances automated security evaluation and lowers the barrier for systematic vulnerability discovery across diverse processor architectures.

Recommended citation: D. R. Dipta, T. Tiemann, B. Gulmezoglu, E. Marin and T. Eisenbarth, "uGen: From Gaps to Exploits — A Multi-Agent, Retrieval-Augmented System for End-to-End Microarchitectural Attack Code Generation." Under review.