Publications
Sort:
Open Access Article Issue
Scaling the Strategy Wall: Efficient Jailbreaking of LLMs via Component-Based Multi-Objective Optimization
Computers, Materials & Continua 2026, 88(3): 43
Published: 23 July 2026
Abstract PDF (2.2 MB) Collect
Downloads:13
Background

Jailbreak attacks, which use crafted prompts to bypass safety alignments of Large Language Models (LLMs) and generate harmful content, pose a significant security threat. Existing methods often optimize for a single objective (e.g., attack success rate), neglecting critical factors like query efficiency, which limits their practicality and generalization.

Methods

We propose a Componentized Multi-Objective Optimization Framework (CMOOF), which introduces a paradigm shift: it searches for generalizable and query-efficient attack strategy templates within a structured, component-based strategy space. CMOOF leverages the NSGA-II algorithm to explicitly co-optimize two first-class objectives: Attack Success Rate (ASR) and Query Efficiency, thereby discovering their Pareto-optimal trade-off frontier.

Results

Experiments on benchmark datasets show significant improvements, with the highest jailbreak success rate reaching 98.75% on models like Llama3, and query efficiency surpassing baselines.

Conclusions

CMOOF redefines jailbreak optimization from instance-level prompt crafting to strategy-level template discovery. The work provides an efficient, scalable, and generalizable jailbreak solution, and the framework offers broader insights for automated red teaming and LLM security defense.

Open Access Review Issue
Beyond Intentions: A Critical Survey of Misalignment in LLMs
Computers, Materials & Continua 2025, 85(1): 249-300
Published: 29 August 2025
Abstract PDF (1.4 MB) Collect
Downloads:174

Large language models (LLMs) represent significant advancements in artificial intelligence. However, their increasing capabilities come with a serious challenge: misalignment, which refers to the deviation of model behavior from the designers’ intentions and human values. This review aims to synthesize the current understanding of the LLM misalignment issue and provide researchers and practitioners with a comprehensive overview. We define the concept of misalignment and elaborate on its various manifestations, including generating harmful content, factual errors (hallucinations), propagating biases, failing to follow instructions, emerging deceptive behaviors, and emergent misalignment. We explore the multifaceted causes of misalignment, systematically analyzing factors from surface-level technical issues (e.g., training data, objective function design, model scaling) to deeper fundamental challenges (e.g., difficulties formalizing values, discrepancies between training signals and real intentions). This review covers existing and emerging techniques for detecting and evaluating the degree of misalignment, such as benchmark tests, red-teaming, and formal safety assessments. Subsequently, we examine strategies to mitigate misalignment, focusing on mainstream alignment techniques such as RLHF, Constitutional AI (CAI), instruction fine-tuning, and novel approaches that address scalability and robustness. In particular, we analyze recent advances in misalignment attack research, including system prompt modifications, supervised fine-tuning, self-supervised representation attacks, and model editing, which challenge the robustness of model alignment. We categorize and analyze the surveyed literature, highlighting major findings, persistent limitations, and current contentious points. Finally, we identify key open questions and propose several promising future research directions, including constructing high-quality alignment datasets, exploring novel alignment methods, coordinating diverse values, and delving into the deep philosophical aspects of alignment. This work underscores the complexity and multidimensionality of LLM misalignment issues, calling for interdisciplinary approaches to reliably align LLMs with human values.

Total 2