Detecting and Correcting Hallucinations in LLM-Generated Code via Deterministic AST Analysis

TL;DR

This paper presents a deterministic static-analysis framework that detects and auto-corrects hallucinations in LLM-generated code by parsing code into ASTs and validating against a dynamically generated knowledge base, achieving high precision and recall.

Contribution

It introduces a novel static-analysis method for detecting and fixing hallucinations in code generated by LLMs, outperforming probabilistic approaches.

Findings

Detected hallucinations with 100% precision and 87.6% recall.

Successfully auto-corrected 77.0% of hallucinations.

Demonstrated reliability of deterministic post-processing for trustworthy code generation.

Abstract

Large Language Models (LLMs) for code generation boost productivity but frequently introduce Knowledge Conflicting Hallucinations (KCHs), subtle, semantic errors, such as non-existent API parameters, that evade linters and cause runtime failures. Existing mitigations like constrained decoding or non-deterministic LLM-in-the-loop repair are often unreliable for these errors. This paper investigates whether a deterministic, static-analysis framework can reliably detect \textit{and} auto-correct KCHs. We propose a post-processing framework that parses generated code into an Abstract Syntax Tree (AST) and validates it against a dynamically-generated Knowledge Base (KB) built via library introspection. This non-executing approach uses deterministic rules to find and fix both API and identifier-level conflicts. On a manually-curated dataset of 200 Python snippets, our framework detected KCHs with 100\% precision and 87.6\% recall (0.934 F1-score), and successfully auto-corrected 77.0\% of all identified hallucinations. Our findings demonstrate that this deterministic post-processing approach is a viable and reliable alternative to probabilistic repair, offering a clear path toward trustworthy code generation.