The field of large language models (LLMs) is rapidly evolving, with a growing focus on addressing security vulnerabilities. Recent research has highlighted the susceptibility of LLMs to various attacks, including jailbreaks, prompt injection, and adversarial manipulations. To mitigate these risks, researchers are exploring innovative solutions, such as AI-driven detection methods, intent-aware fine-tuning, and dual-track defense frameworks. These approaches aim to enhance the robustness and safety of LLMs, ensuring their reliable deployment in critical applications. Noteworthy papers in this area include: Enhancing GraphQL Security by Detecting Malicious Queries Using Large Language Models, which presents a novel AI-driven approach for real-time detection of malicious GraphQL queries. Mitigating Jailbreaks with Intent-Aware LLMs, which proposes a simple and lightweight fine-tuning approach to improve LLM robustness against jailbreak attacks.
Advances in Large Language Model Security
Sources
Enhancing GraphQL Security by Detecting Malicious Queries Using Large Language Models, Sentence Transformers, and Convolutional Neural Networks
Mitigating Jailbreaks with Intent-Aware LLMs
MAJIC: Markovian Adaptive Jailbreaking via Iterative Composition of Diverse Innovative Strategies
Too Easily Fooled? Prompt Injection Breaks LLMs on Frustratingly Simple Multiple-Choice Questions
Quantifying Loss Aversion in Cyber Adversaries via LLM Analysis
Involuntary Jailbreak
Special-Character Adversarial Attacks on Open-Source Language Model
CCFC: Core & Core-Full-Core Dual-Track Defense for LLM Jailbreak Protection
Online Incident Response Planning under Model Misspecification through Bayesian Learning and Belief Quantization
Universal and Transferable Adversarial Attack on Large Language Models Using Exponentiated Gradient Descent
SDGO: Self-Discrimination-Guided Optimization for Consistent Safety in Large Language Models