今日速览
| 序号 | 标题 | 来源 | 日期 | 主题 | 推荐等级 |
|---|---|---|---|---|---|
| 1 | AfterDeleteBench: Auditing Post-Deletion Information Persistence in LLM Agents | arXiv | 2026-10-22 | RAG | 高 |
| 2 | Muffin - A Shared Memory Buffer for Asynchronous AI Agents | arXiv | 2026-10-22 | AI-Agent | 高 |
重点论文与技术动态
1. AfterDeleteBench: Auditing Post-Deletion Information Persistence in LLM Agents
- 来源：arXiv
- 日期：2026-10-22
- 作者/机构：Ehsan Saleh, Roozbeh Nahavandi, Sama Salarian, Raef Bassily, Andrew Perrault
- 主题标签：
RAG,arXiv - 推荐等级：高
- 分类：cs.CR, cs.CL, cs.LG
一句话结论
AfterDeleteBench评测表明，LLM智能体在删除原始记忆后仍可通过衍生状态恢复机密，证明记忆删除不能仅作为存储操作，需从系统行为层面综合评估。
核心内容
- 信息残留显著：单次查询即可从Qwen2.5-7B中恢复约2/3已删除机密，而从未接触的对照组恢复率为零。
- 支持非逐字重构：即使无完整原文记录，仅凭分散组件与规则，Qwen、Mistral和Phi-3的精确重构成功率仍达62%–99%。
- 防御效果差异：溯源感知闭包可消除测试中的恢复现象；而字面写入过滤仅能阻断重新写入，无法清除已存残留。
方法与数据
- 基于合成机密构建受控行为审计基准，对比“接触后删除”与“从未接触”对照组，评估Qwen、Mistral与Phi-3等固定权重智能体。
价值判断
- 值得关注：揭示了摘要等衍生状态会导致底层存储删除失效的安全盲区。
- 可复用点：通过受控对照排除模型预训练先验的记忆审计与因果归因设计。
- 局限/待核查：真实多轮交互场景下的残留情况及闭源商业模型的表现摘要未明确。
摘要
Many LLM agents maintain external persistent memory and derive additional state from past interactions, such as summaries. Deleting an original memory record therefore need not remove information carried by its descendants. We introduce AfterDeleteBench, a controlled behavioral audit for this failure mode in fixed-weight LLM agents. For each synthetic secret, we compare an exposed-then-deleted agent with an otherwise matched agent that was never exposed to the secret, allowing post-deletion recovery to be attributed to the original exposure. In our primary setting, a single query recovers the deleted secret from Qwen2.5-7B-Instruct in roughly two thirds of cases, while matched never-exposed controls have zero recovery. Crucially, recovery does not require a surviving verbatim copy: when the complete secret string is absent and only separately stored components remain under a public composition rule, exact reconstruction succeeds in 62-99% of cases across Qwen, Mistral, and Phi-3. We also show that the response interface is part of the measurement system: target-bearing replies can be scored as failures when the requested format and parser are misaligned. Finally, we separate reconstruction from write-back and fresh-session recall. Provenance-aware closure eliminates observed recovery in the tested positive-baseline settings, whereas an exact-literal write filter blocks reinsertion but cannot erase information already present. Together, these results show that source deletion should be evaluated behaviorally at the level of the agent and its memory system, not only as a storage operation.2. Muffin - A Shared Memory Buffer for Asynchronous AI Agents
- 来源：arXiv
- 日期：2026-10-22
- 作者/机构：Ping-Lin Kuo, Suman Navaratnarajah, Christian Frey, Gresa Shala, Josif Grabocka
- 主题标签：
AI-Agent,arXiv - 推荐等级：高
- 分类：cs.AI, cs.MA
一句话结论
Muffin 提出了一种基于共享内存缓冲区的异步多智能体交互协议，在保持私有推理本地化的同时实现中间发现的异步交换，显著提升了推理性能并降低了计算成本。
核心内容
- 针对现有同步轮流辩论中交互低效的问题，构建了异步多智能体协作框架。
- 引入协调器按推理需求路由任务，仅将复杂查询分发给多智能体系统。
- 智能体通过共享内存缓冲区异步交换中间发现，私有推理保留在本地，无需全轨迹共享。
方法与数据
- 提出协调器路由结合共享内存缓冲区的异步推理机制；在 20 个基准设置中进行了对比实验，具体数据集名称摘要未明确。
价值判断
- 值得关注：在 17/20 个基准设置中取得最高准确率；相比多智能体辩论，准确率提升 19.2 个百分点，同时减少了 77.4% 的 Token 消耗。
- 可复用点：按需分流的协调器设计，以及“私有推理本地化 + 中间结果共享”的缓冲区通信范式。
- 局限/待核查：具体评估的基准任务类型及协调器的路由实现细节在摘要中未明确。