Symbolic execution is effective at uncovering security violations, but it suffers from path explosion: the number of paths to explore quickly becomes unmanageable. Tools such as KLEE cope by using path prioritization heuristics to decide which states to explore first, usually optimizing for code coverage.
A new arXiv paper, "Directed Symbolic Execution for Vulnerability Discovery: An LLM-Guided Approach in KLEE," examines using a language model to guide this process. The title points to a directed strategy aimed at vulnerability discovery, rather than broad coverage-oriented exploration.
The provided abstract only lays out the core problem and the standard heuristic solution; it does not describe the LLM guidance mechanism in detail. Still, the work signals an ongoing effort to make symbolic execution more practical by combining it with learned guidance.