三篇新研究论文探讨了大型语言模型 (LLM) 和大型音频语言模型 (LALM) 的漏洞和防御。第一篇论文详细介绍了音频越狱攻击和防御的分类法，强调目前的防御措施通常会牺牲可用性来换取鲁棒性。第二篇论文全面回顾了 LLM 的漏洞，对攻击和防御进行了分类，并指出了在弹性对齐和自动检测等领域的研究空白。第三篇论文介绍了“越狱规模定律”，证明了对抗性提示如何将攻击成功率从多项式增长转变为指数增长，这种现象在各种 LLM 和攻击方法中都有观察到。 AI
影响 新研究强调了 LLM 和 LALM 安全方面不断升级的风险，并强调需要更强大、更易于使用的防御措施来应对复杂的越狱技术。
排序理由 该集群包含三篇学术论文，详细介绍了对 LLM 和 LALM 漏洞及防御的研究。
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
arXiv:2605.30031v1 Announce Type: cross Abstract: Large Audio Language Models (LALMs) expand jailbreak risks from token-level prompting to the full speech perception-to-reasoning pipeline, where unsafe behavior can be induced through semantics, acoustic style, signal artifacts, o…
arXiv:2410.15236v4 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have transformed artificial intelligence by advancing natural language understanding and generation, enabling applications across fields beyond healthcare, software engineering, and conversatio…
arXiv:2603.11331v3 Announce Type: replace-cross Abstract: Adversarial attacks can reliably steer safety-aligned large language models toward unsafe behavior. Empirically, we find that adversarial prompt-injection attacks can amplify attack success rate from the slow polynomial gr…
Large Audio Language Models (LALMs) expand jailbreak risks from token-level prompting to the full speech perception-to-reasoning pipeline, where unsafe behavior can be induced through semantics, acoustic style, signal artifacts, or internal representations. Existing work studies …