Effective and scalable vulnerability detection in stripped binaries

Develop effective and scalable techniques to detect security vulnerabilities in stripped binary files.

Background

The paper targets vulnerability detection directly from compiled binaries, particularly stripped binaries where symbol and debug information are removed. Traditional reverse engineering pipelines (disassembly and decompilation) lose high-level semantic information, making automated detection difficult and often reliant on expert human analysis.

While LLMs have recently improved the readability of decompiled code, the authors note that moving from human-readable recovery to reliable, scalable vulnerability detection at the binary level remains unresolved. The work introduces Vul-BinLLM as a step toward this goal, but explicitly states that achieving effective and scalable detection in stripped binaries is an open problem.