Dr.Binary: Binary Analysis Built for AI Agents
COM2 Level 2
MR3, COM2-02-26
Abstract:
Recent advances in LLMs and agentic frameworks have demonstrated impressive capabilities in cybersecurity. However, simply equipping AI agents with existing binary-analysis tools does not necessarily lead to effective and reliable analysis. In this talk, I will introduce Dr.Binary, a binary-analysis platform designed specifically for AI agents. I will discuss three key designs: a semantic workspace that makes binaries accessible and searchable by agents, interactive and scripted concolic reasoning that combines debugging with symbolic execution, and agent-oriented binary diffing. Through concrete use cases and benchmark results, I will demonstrate how specialized analysis infrastructure can significantly enhance LLM capabilities while reducing analysis costs. I will conclude with lessons learned and open research challenges in agentic binary analysis.
Bio:
Heng Yin is a Professor in the Department of Computer Science and Engineering at the University of California, Riverside, and the co-founder of Deepbits Technology. His research focuses on cybersecurity, binary analysis, program analysis, and software security. He has led the development of several widely recognized research systems, including DECAF, DroidScope, and SymFit. His recent work explores the intersection of AI and program analysis, particularly the use of AI agents for automated binary analysis and vulnerability discovery. Through Deepbits, he is also leading the development of Dr.Binary, an AI-native binary-analysis platform that brings advanced program-analysis capabilities to AI agents.