今日速览
| 序号 | 标题 | 来源 | 日期 | 主题 | 推荐等级 |
|---|---|---|---|---|---|
| 1 | Agent Against Agent: An Agentic System for Automatic Prompt Injection Red Teaming | arXiv | 2026-08-05 | AI-Agent | 高 |
| 2 | CoPlan: A Trustworthy Co-Intelligence Interface for Care Planning through Role-Based Contestable Argument Graphs | arXiv | 2026-08-05 | AI-Agent | 高 |
重点论文与技术动态
1. Agent Against Agent: An Agentic System for Automatic Prompt Injection Red Teaming
- 来源:arXiv
- 日期:2026-08-05
- 作者/机构:Yanting Wang, Chenlong Yin, Runpeng Geng, Jinyuan Jia
- 主题标签:
AI-Agent,arXiv - 推荐等级:高
- 分类:cs.CR
一句话结论
PIMiner通过构建可迁移策略库,实现提示注入红队测试的高效泛化,显著提升对多种LLM的攻击成功率。
核心内容
- 现有提示注入红队方法依赖强化学习,泛化性差,难以适应新目标LLM。
- PIMiner训练时在(数据集, 目标模型)序列上构建策略库,测试时可直接迁移至未见过的LLM,无需额外训练。
- 实验表明,在IPIArena和AgentDojo基准上,对Gemini-2.5-Pro等模型达到高攻击成功率(如76.2% ASR)。
方法与数据
- 方法:基于序列的(数据集, 目标模型)对训练构建策略库。
- 数据:测试基于IPIArena和AgentDojo基准。
价值判断
- 值得关注:有效解决现有方法泛化性问题,提升红队测试效率。
- 可复用点:策略库可直接用于新LLM,无需重新训练。
- 局限/待核查:摘要未明确。
摘要
Prompt injection poses significant security risks to LLM agents. Efficient and effective red-teaming is therefore critical, both for evaluating these risks and for collecting training data to improve defenses. Existing state-of-the-art prompt injection red-teaming methods primarily rely on reinforcement learning (RL), producing attacker models that often generalize poorly to new target LLMs. In this work, we develop PIMiner, an agentic system for prompt injection red-teaming. During training, PIMiner is trained on a sequence of (dataset, target model) pairs and builds a strategy library from scratch. At test time, the learned strategy library can be directly transferred to a previously unseen target LLM without additional training. PIMiner requires only a small number of queries to a target agent (e.g., 10) per test sample. Experimental results demonstrate that PIMiner achieves strong performance. On IPIArena, it attains a 76.2% ASR against Gemini-2.5-Pro, 61.9% ASR against GPT-5.1, and 42.9% ASR against Claude-Sonnet-4.5. On AgentDojo, it achieves an 86.7% ASR against Gemini-2.5-Pro, 53.3% ASR against GPT-5.1, and 40.0% ASR against Claude-Sonnet-4.5.2. CoPlan: A Trustworthy Co-Intelligence Interface for Care Planning through Role-Based Contestable Argument Graphs
- 来源:arXiv
- 日期:2026-08-05
- 作者/机构:Hung Truong Thanh Nguyen, Hélène Fournier, Piper Jackson, Makoto Itoh, Shannon Freeman, Rene Richard
- 主题标签:
AI-Agent,arXiv - 推荐等级:高
- 分类:cs.AI, cs.MA, cs.SE
一句话结论
CoPlan通过角色化可争议论点图实现可信的协作智能护理规划界面。
核心内容
- CoPlan解决AI系统固定输出问题,使临床医生、患者等利益相关者能在临床判断、患者价值观或现实可行性冲突时检查、挑战和修改护理计划。
- 采用多代理工作流:AI代理生成候选干预及支撑/挑战论点,人类规划者可操作论点(接受、拒绝、修改或添加),再生成最终计划。
- 结合协作智能(人类与AI互补专家)与可争议性(推荐开放检查、修改、证明),保留人类代理和临床问责。
方法与数据
- 在老龄化居家护理规划场景中演示系统功能。
- 摘要未明确具体数据集或实验细节。
价值判断
- 值得关注:显著提升AI支持护理规划的可信度与人类参与度,避免固定输出导致的决策僵化。
- 可复用点:可争议界面设计框架可推广至其他医疗决策场景,强化人机协作。
- 局限/待核查:摘要未明确提及局限性或验证范围。