提出DualView防御机制，有效阻止针对个人AI代理的间接提示注入攻击，开创性安全方案。
arXiv:2607.03821v1 Announce Type: cross Abstract: Personal AI agents that run on the user's local machine, such as OpenClaw, automate daily tasks incl…
这篇论文发现对齐机制能有效防御大语言模型属性推断攻击，保护数据集隐私，为模型安全提供新思路。
arXiv:2606.10217v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly fine-tuned on domain-specific datasets that may contain …
大模型后训练阶段面临全新威胁：顺序数据投毒攻击，揭示安全漏洞与防御方法。
arXiv:2606.04929v1 Announce Type: new Abstract: LLM post-training proceeds through multiple stages, e.g., supervised fine-tuning (SFT) followed by rei…
针对语言模型训练中突发性错位问题，提出在训练阶段进行防御的新方法，来自arXiv最新预印本。
arXiv:2508.06249v3 Announce Type: replace-cross Abstract: Fine-tuning lets practitioners repurpose aligned large language models (LLMs) for new domain…
动态红队测试与集成感知防御：首个覆盖Gmail、Salesforce等主流SaaS集成的LLM Agent安全基准，揭示间接提示注入的真实威胁。
arXiv:2606.02240v1 Announce Type: cross Abstract: Indirect prompt injection in tool-use agents is a concrete production threat: LLM agents read from i…
提示注入是AI代理最致命的漏洞，研究表明现有防御手段可能永远无法彻底防范
arXiv:2605.17634v1 Announce Type: cross Abstract: Prompt injection is the most critical vulnerability in deployed AI agents. Despite recent progress, …
针对大模型多智能体系统的传染性越狱攻击，提出前瞻性防御方法，在感染扩散前阻断威胁。
arXiv:2605.01758v3 Announce Type: replace Abstract: Large multimodal model-based Multi-Agent Systems (MASs) enable collaborative complex problem solvi…