EAGER: Edge-Aligned LLM Defense for Robust, Efficient, and Accurate Cybersecurity Question Answering

TL;DR

EAGER is a novel framework that enhances cybersecurity question answering with edge-efficient, robust, and accurate LLMs by combining quantization and domain-specific preference alignment, suitable for edge devices.

Contribution

It introduces a unified approach integrating quantization and preference alignment for robust, efficient edge deployment of cybersecurity LLMs, using QLoRA and DPO without human labels.

Findings

Reduces adversarial attack success rates by up to 7.3x.

Improves QA accuracy by up to 55%.

Achieves lowest latency on Jetson Orin.

Abstract

Large Language Models (LLMs) are highly effective for cybersecurity question answering (QA) but are difficult to deploy on edge devices due to their size. Quantization reduces memory and compute requirements but often degrades accuracy and increases vulnerability to adversarial attacks. We present EAGER, an edge-aligned defense framework that integrates parameter-efficient quantization with domain-specific preference alignment to jointly optimize efficiency, robustness, and accuracy. Unlike prior methods that address these aspects separately, EAGER leverages Quantized Low-Rank Adaptation (QLoRA) for low-cost fine-tuning and Direct Preference Optimization (DPO) on a self-constructed cybersecurity preference dataset, eliminating the need for human labels. Experiments show that EAGER reduces adversarial attack success rates by up to 7.3x and improves QA accuracy by up to 55% over state-of-the-art defenses, while achieving the lowest response latency on a Jetson Orin, demonstrating its practical edge deployment.