SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement

Abstract

Despite the significant efforts devoted to aligning large language models (LLMs) with human values and ensuring safe deployment, recent work has revealed that LLMs remain vulnerable to adversarial jailbreak attacks that can bypass safety guardrails and elicit harmful responses. Many defense methods are proposed to detect jailbreaks but they are limited in their effectiveness to counter wide-range optimization-based jailbreak mechanisms that can yield highly fluency-optimized or harmful semantic

Categories

Framework mappings

Suggested from the entry's categories.

@misc{vo2026safeguard,
  title = {{SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement}},
  author = {Quoc Viet Vo and Trung Le and Damith C. Ranasinghe and Ehsan Abbasnejad},
  year = {2026},
  month = sep,
  eprint = {2609.05850},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.05850}
}