Work

Circuit-Aware LoRA: Reducing Hallucinations in Security LLMs

Year: 2026

Type: article

Abstract: Large Language Models (LLMs) fine-tuned for cybersecurity tasks suffer from a critical hallucination problem: they often claim to detect vulnerabilities with high confidence when no real evidence exists. We present Circuit-Aware LoRA, a targeted fine-tuning method that reduces hallucinations by strengthening concept support in semantically-weak transformer layers.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.21537141

Author: Yahya Saqban

Language: English

Topic: Adversarial Robustness in Machine Learning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7170843854

FWCI:

Cites:

Cited by:

Related to: