ClearAgent: Agentic Binary Analysis for Effective Vulnerability Detection
summary
ClearAgent builds an agentic binary-analysis pipeline that wires LLMs to a binary language server and a context manager, using a symbolic lifter (plankton) plus Ghidra-style tools to lift binaries to disassembly, IR, and decompiled pseudocode. It iteratively queries those formats, finds 19 CTF-style bugs with comparable results to state-of-the-art agents, reproduces a historical IoT CVE by synthesizing a concrete exploit input, and emits PoC scripts; remaining failure modes include inaccurate IR and unresolved indirect-call targets.
AI-written summary · about
chapters (11)
- 0:00ClearAgent project intro
- 1:26Binary code analysis challenges
- 2:29Agentic paradigm for analysis
- 3:32Limitations of existing agents
- 5:16ClearAgent framework overview
- 6:21Binary language server design
- 8:07Context manager for exploration
- 9:35CTF challenges evaluation
- 11:23CVE reproduction case study
- 12:30Future directions
- 13:11Q&A
related talks
31mA Dummy's Guide to Agentic Exploit Generation
27mKeynote: Towards Building Safe and Secure Agentic AI
33mAgentic ProbLLMs: Exploiting AI Computer-Use and Coding Agents
29mBuilding a Hackbot: How Ethiack's Hackian Finds Real Vulnerabilities
36mAgentic AI Under Attack: Live Demos of Exploits Through Autonomy and Trust
19mKeynote: Towards Building Safe and Secure Agentic AI
34mSecure Your Coding Agent Like It’s Malware
39mThe AI Arms Race in Vulnerability Management, Who’s Winning?