Cutting the Root of Hallucination: Structural Trimming for Vulnerability Mitigation in Code LLMs

Yage Zhang

Keywords: LLM hallucinations, code generation, program repair, vulnerability mitigation, structural pruning, abstract syntax tree, hallucination detection, CSHS, model-agnostic risk estimation, generative code safety

TL;DR: LLMs hallucinate code often with security risks. We trace these structurally, prune them surgically, and predict repair effectiveness. Our method patches code and mitigates risk using a transferable score (CSHS).

Abstract: We introduce a structural perspective on hallucinations in code-generating language models, framing them as causality anchors in syntax graphs that trigger cascading semantic errors and latent security flaws. This work is the first to systematically connect code hallucinations with vulnerability risks, offering a unified conceptual and practical framework to address them. At the heart of our approach is the notion of hallucination anchors, localized subtrees in the abstract syntax tree (AST) that serve as root causes of defective logic. We propose Structural Trimming (ST), a targeted mitigation method that removes these anchors while preserving functional semantics. To anticipate the effect of trimming, we introduce the Compositional Structural Hallucination Score (CSHS), which quantifies the likelihood that pruning will improve robustness. By grounding error reduction in the syntax graph itself, our method reframes hallucination mitigation as a structured intervention process interpretable, generalizable, and actionable.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the COLM Code of Ethics on https://colmweb.org/CoE.html

Author Guide: I certify that this submission complies with the submission instructions as described on https://colmweb.org/AuthorGuide.html

Submission Number: 269