Defusing Logic Bombs in Symbolic Execution with LLM-Generated Ghost Code

TL;DR

Gordian is a hybrid symbolic execution framework that uses LLM-generated ghost code to improve solver effectiveness and coverage in analyzing complex, real-world programs, significantly outperforming prior methods.

Contribution

This work introduces Gordian, which selectively employs LLMs to generate ghost code that enhances symbolic execution's ability to handle solver-hostile code fragments and unbounded structures.

Findings

Gordian improves coverage by 52-84% over traditional symbolic execution.

Gordian outperforms pure LLM-based techniques by 86-419% in coverage.

Gordian reduces LLM token usage by 90-96%.

Abstract

Symbolic execution is a powerful program analysis technique, but its effectiveness is fundamentally limited by solver-hostile program fragments, complex numerical reasoning, and unbounded heap structures. Recent work proposed replacing constraint solvers with large language models (LLMs) to bypass these limitations, but such approaches struggle to analyze real-world codebases, where deep execution paths require globally consistent reasoning across many interacting constraints. We present Gordian, a hybrid symbolic execution framework that uses LLMs selectively to generate lightweight ghost code that aids an SMT solver in handling solver-hostile code fragments, while preserving its precise, global reasoning capability. In particular, we propose three types of ghost code: (1) inversion of difficult code fragments with iterative bidirectional constraint propagation, (2) modeling via solver-friendly surrogates while preserving relevant behavior, and (3) semantic partitioning of unbounded heap spaces. We implemented Gordian on top of the KLEE symbolic execution engine and evaluated it on synthetic "logic bombs" capturing distinct symbolic reasoning challenges, a popular mathematical library FDLibM, and three structured-input programs (libexpat, jq, and bc). Across all benchmarks, Gordian improves coverage on average by 52-84% over traditional symbolic execution baselines, and by 86-419% over LLM-based techniques, while reducing LLM token usage by an average of 90-96%. This highlights the practicality and effectiveness of this approach in real-world settings.