论文

本综述从数据中心化视角梳理 LLM 智能体的隐私问题，围绕智能体所接触的数据而非按攻击类型组织领域，并以"data agent"作为处理数据的 LLM 智能体的简称。This survey surveys the privacy of LLM agents from a data-centric view, organizing the field around the data an agent touches rather than by attack type, and uses data agent as shorthand for an LLM agent that works with data.

PAVE 是一种逐步式标注协议，利用来自故障注入的已知干预来重建因果传播路径；逐步式的因果真值正是可信的基于 LLM 的 RCA Agent 所缺失的关键一环。PAVE, a step-wise labeling protocol that leverages known interventions from fault injection to reconstruct causal propagation paths, is introduced, a step-wise causal ground truth is the missing piece for trustworthy LLM-based RCA agents.

本工作将带外防御组织为经典完整性保护、引用监控与最小权限的具体实例，对它们覆盖与未覆盖的内容进行结构化对比；与该假设一致但尚未被证实的是：确定性的带外强制执行相比带内检测，是更难被自适应攻击者攻破的目标。This work organizes out-of-band defenses as instances of classical integrity protection, reference monitoring, and least privilege, yielding a structured comparison of what they do and do not cover, consistent with, but not established, the hypothesis that deterministic out-of-band enforcement is a harder target for an adaptive attacker than in-band detection.

一项关于 LLM 程序修复中执行行为的两阶段实证研究表明，当前 Agent 不加区分地使用执行，在收益甚微的实例上仍付出其代价，因此应被视为具有显式成本-收益权衡的资源。A two-stage empirical study of execution behavior in LLM-based program repair suggests that current agents apply execution indiscriminately, paying its cost on instances where it provides little benefit, and should be treated as a resource with an explicit cost-benefit tradeoff.

路由、投票、级联、融合与 Mixture-of-Agents 等多模型 LLM 系统常被用于超越单模型精度；研究表明其增益受限于一个该领域鲜少报告的量化指标，且在缺乏强查询级路由信号时，组合模型很少能胜过单一最佳模型。Multi-model LLM systems such as routing, voting, cascades, fusion, and mixture-of-agents are used to beat single-model accuracy, it is shown that their gain is capped by a quantity the field rarely reports, and combining models rarely beats the single best model without a strong query-level routing signal.

本文认为，将上下文学习（ICL）扩展至终身设置是 AI Agent 持续学习的实用方案；要在固定硬件预算下用 Transformer 理解终身上下文，需要注意力的参数化形式。It is argued that extending in-context learning to lifelong settings is a practical solution for continual learning in AI agents and that parametric forms of attention are needed to understand a lifetime of context with transformers on a fixed hardware budget.

本文展示了用于 IT-Grundschutz（IT-GS）认证部分自动化的多 Agent 系统（MAS）架构结合混合检索增强生成（HybridRAG）的技术实现与实证评估，并为强化合规严谨性引入两项新的 MAS 架构技术贡献。This paper presents the technical implementation and empirical evaluation of a Multi-Agent System (MAS) architecture combined with Hybrid Retrieval Augmented Generation (HybridRAG) for the partial automation of IT-GS certification and introduces two novel technical contributions to the MAS architecture to enforce the compliance rigor.

研究发现，强化学习（RL）与监督微调（SFT）交错训练可显著提升稳定性，但在格式与内容分布外（OOD）评测下性能下降；并展示了多样化监督信号如何引导探索式学习。It is found that interleaving supervised fine-tuning with RL substantially improves stability, but exhibits degraded performance under format and content out-of-distribution (OOD) evaluation, and how diverse supervisory signals can guide exploratory learning is demonstrated.

结果表明，工具使用评测应从函数调用准确率转向不可靠工具环境下的任务完成度，并建议工具使用评测应从函数调用准确率转向不可靠工具环境下的任务完成度。（注：原文末句疑似重复）Results suggest that tool-use evaluation should move beyond function-call accuracy toward task completion under unreliable tool environments, and suggest that tool-use evaluation should move beyond function-call accuracy toward task completion under unreliable tool environments.

提出 Transparent Two-Pass Execution，一种在推理时将工具执行与 schema 约束响应生成解耦的策略；实验结果表明该方法无需模型重新训练即可恢复工具调用能力，同时保持结构化输出保证。Transparent Two-Pass Execution is proposed, an inference-time strategy that decouples tool execution from schema-constrained response generation and experimental results show that this approach restores tool invocation while preserving structured output guarantees without requiring model retraining.

提出 SkillHone，一个基于持久决策历史实现 Agent Skill 持续进化的 harness；在内部工具辅助的分析场景中提升了准确率，并在未预先集成搜索栈的情况下优于商业支持的深度研究 Agent。SkillHone is introduced, a harness for continual agent skill evolution grounded in persistent decision history that improves accuracy on internal tool-mediated analysis scenarios and outperforms commercially backed deep-research agents without a pre-integrated search stack.

提出 conversational infill：让一个小型 talker 模型在外部 reasoner 模型产生结果前即时生成上下文相关的回复以掩盖延迟，并在推理过程中将 reasoner 流式输出的知识流畅地融合到回复中。Conversational infill is introduced, where a small talker model both immediately generates contextually grounded responses to hide the latency of an external reasoner model and fluently integrates streamed reasoner knowledge into its responses during inference.

AOHP 的核心设计原则是将 Agent 视为 OS 中的一等公民，从而支持自适应用户界面以及对 Agent 友好的运行时环境；在任务完成度、执行成本和安全策略合规性方面均展现出明显优势。The core design principle of AOHP is to treat agents as first-class OS actors, enabling adaptive user interfaces and agent-friendly runtime environments, and shows clear advantages in task completion, execution cost, and security-policy compliance.

本研究收集了 9,041 个使用流行 AI Agent 开发的开源应用，审计了 200 个公开部署的应用，发现 1,186 个漏洞，并对 vibe-coded 应用的安全态势提供了实证分析。This study collects 9,041 open-source applications developed using popular AI agents, audits 200 publicly deployed applications, uncovering 1,186 vulnerabilities, and provides an empirical understanding of the security landscape of vibe-coded applications.

对话式与工具使用的 LLM Agent 在上下文窗口中同时从多个方向被填充，而必须在多轮之间回忆信息的 Agent（即 memory 的典型场景）恰恰是 recency 截断失效的地方。Conversational and tool-using LLM agents operate over a context window that fills from several directions simultaneously, and agents that must recall information across many turns, the defining case for memory, are precisely where recency truncation fails.

提出 S-Agent，一种面向连续多视图图像与视频理解与推理的空间工具使用 Agent 范式，以无需训练的方式持续提升开源与闭源 VLM。This work introduces S-Agent, a spatial tool-use agentic paradigm for understanding and reasoning over continuous multi-view images and videos, and consistently improves both open-source and closed-source VLMs in a training-free manner.

揭示了指导的生成方式才是决定性变量，并提出 probe-and-refine tuning（探测-微调）：通过合成 bug 修复探测任务，利用单次 LLM 调用迭代诊断并修补仓库的指导文件，调优过程中不涉及 Agent 循环或工具调用。It is shown that how the guidance is produced is the decisive variable, and probe-and-refine tuning is introduced, a procedure that uses synthetic bug-fix probes to iteratively diagnose and patch a repository's guidance file through single-shot LLM calls, with no agent loop or tool use during tuning.

提出一种特权感知的训练后防御方法，教导 Agent 优先选用足够的低权限工具，仅在必要时升级；该方法在保留通用能力的同时大幅减少了不必要的高权限工具使用。A privilege-aware post-training defense that teaches agents to prefer sufficient lower-privilege tools and escalate only when necessary is introduced, showing that this defense substantially reduces unnecessary high-privilege tool use while preserving general capabilities.

提出 Qwen-RobotNav，一种基于 Qwen-RobotNav 构建的可扩展导航模型，通过参数化接口沿两个互补维度应对该问题：用于选择导航行为的多种任务模式，以及用于控制视觉历史编码方式的可控观测参数。Qwen-RobotNav is presented, a scalable navigation model built on Qwen-RobotNav that addresses it through a parameterised interface with two complementary dimensions: multiple task modes that select the navigation behaviour, and controllable observation parameters that govern how visual history is encoded.

提出 GateMem，一个面向多主体共享内存 Agent 的基准，联合评估合法长程请求及其状态更新的效用、跨上下文授权边界的访问控制，以及 Agent 在收到显式删除请求后的主动遗忘能力。GateMem is introduced, a benchmark for multi-principal shared-memory agents that jointly evaluates utility for legitimate long-horizon requests with state updates, access control across contextual authorization boundaries, and agent-facing active forgetting after explicit deletion requests.

C-Trace 提出一个验证框架，将 GDPR 的部分要求（包括同意、目的限制、数据最小化与被遗忘权）表达为针对 Agent 执行轨迹的正式策略谓词。C-Trace (Compliance Trace based Runtime Agent Conformance Enforcement), a verification framework that expresses a subset of GDPR requirements, including consent, purpose limitation, data minimization, and the right to erasure, as formal policy predicates over agent execution traces is proposed.

RODS（Reward-driven Online Data Synthesis）通过将进度奖励方差重新用作零成本边界检测器，在 RL 训练与数据生成之间形成闭环，无需在训练已有的 rollout 之外增加额外推理。RODS (Reward-driven Online Data Synthesis) closes the loop between RL training and data generation by repurposing the progress reward variance as a practical, zero-cost boundary detector that requires no extra inference beyond the rollouts already computed for training.

对涵盖前沿闭源与开源模型、共 22 个模型在多个规模上的评估发现，所有模型家族均存在不可忽略的隐私泄露，且指令遵循能力与泄露率呈正相关。Evaluating 22 models spanning frontier proprietary and open-source models at multiple scales, it is found that all model families exhibit non-trivial leakage, and that instruction- following ability correlates with leakage rate.

该工作以模拟的共享工作空间人机团队为受控实验环境，研究协作结构如何影响团队行为，并表明协调结构是决定可用能力能否提升团队结果的关键。This work uses simulated shared-workspace human-AI teams as a controlled testbed for studying how collaboration structure shapes team behavior, and suggests that coordination structure is central to whether available capability improves team outcomes.

结果表明，在基于 MCP 的 Agent 中，来源归因是事实性验证的一个独立维度；提出 ProvenanceGuard，一种针对 MCP 依据回答的来源感知验证器。Results show that source attribution is an independent axis for factuality verification in MCP-based agents, and ProvenanceGuard, a source-aware verifier for MCP-grounded answers is introduced.

本文介绍 Cordon，一个事务性运行时系统，用于在提交前暂存并验证 Agent 的不可逆操作；其在保持良性任务完成的同时降低不可逆操作失败率，且仅带来适度的审批与时延开销。This paper introduces Cordon, a transactional runtime system for staging and validating irreversible agent effects before commit and reduces irreversible-effect failures while preserving benign task completion with modest approval and latency overhead.

结果表明模型倾向于选择有害场景，在中性预订选项上的表现低于随机猜测水平，其中 Claude 4.8 取得最高分 64.7%。The results indicate that models tend to prefer harmful scenarios, performing below the random chance rate for selecting a neutral booking option, with Claude $4.8$ achieving the highest performance at $64.7\%$.

提出一种多 Agent 框架，通过以确定性编排约束替代 "LLM-as-a-judge" 路由，解决可能在到达患者前未被发现的过早诊断交接与静默临床幻觉问题；观察到 OLDCARTS 完整度与语义熵之间存在统计显著的负相关，提示结构化信息采集与诊断不确定性降低相关。A multi-agent framework that addresses premature diagnostic handoff and silent clinical hallucinations that may go undetected before reaching the patient by replacing ``LLM-as-a-judge''routing with deterministic orchestration constraints is proposed and observes a statistically significant negative correlation between OLDCARTS completeness and semantic entropy, suggesting that structured information gathering is associated with reduced diagnostic uncertainty.

提出一种针对 LLM Agent 的自动化提示优化框架，将"观测到动作"流水线分解为目标条件描述子 Agent 与动作选择 Agent，并通过由 LLM 驱动、以环境回报为指导的进化循环迭代优化各模块提示。An automated prompt optimization framework for LLM agents that decomposes the observation-to-action pipeline into a goal-conditioned descriptor agent and an action selection agent, and iteratively refines each module's prompt through an LLM-driven evolutionary loop guided by environment returns is introduced.

本文认为，Agentic AI 的近期价值不在于完全自主或削减人力，而在于面向简单与中等复杂度业务流程的受控部分自主。It is argued that the near term value of Agentic AI does not lie in full autonomy or workforce reduction, but in controlled partial autonomy for simple and medium complexity business processes.

本文引入 trace-economic underwriting，将工具调用 trace 映射为客户风险敞口与可索赔损失，并以此表示用于定价、控制与风险转移，使用确定性经济标签而非 LLM 评判器。T trace-economic underwriting is introduced, which maps tool-use traces to customer exposure and claimable loss, then uses this representation for pricing, control, and risk transfer, and uses deterministic economic labels rather than an LLM judge.

新加坡 AI Safety Institute 与韩国 AI Safety Institute 联合评估了涵盖客服、DevOps、网页自动化以及企业与个人生产力场景下 12 项真实非对抗任务中的 Agent 数据泄露问题，表明操作性数据泄露是与对抗性数据外泄不同的一阶 Agent 安全问题。A joint evaluation by the Singapore AI Safety Institute and the Korea AI Safety Institute examining agent data leakage in 12 realistic, non-adversarial tasks spanning customer support, DevOps, web automation, and enterprise and personal productivity indicates that operational data leakage is a first-order agent-safety concern distinct from adversarial exfiltration.

对 Agent 行为的分析揭示了长视野经济交互中的显著差异：表现更好的模型与其他企业的沟通更为活跃，而 Claude Haiku 4.5 则表现出 idle-drift 失效模式，在生成连贯评估与规划的同时仍反复选择不行动。Analysis of agent behavior reveals substantial differences in long-horizon economic interaction: higher-performing models communicate more actively with other firms, whereas Claude~Haiku~4.5 exhibits an idle-drift failure mode, repeatedly choosing inaction despite producing coherent assessments and plans.

AgentOdyssey 被提出，这是一个新颖的评估框架，通过程序化方式生成包含丰富实体、世界动态和长视野任务的开放式文本游戏，并发现短期记忆对多种智能体范式均有益，是智能体测试时训练的重要组成部分。AgentOdyssey is introduced, a novel evaluation framework that procedurally generates open-ended text games with rich entities, world dynamics, and long-horizon tasks and finds that short-term memory benefits multiple agent paradigms and is an important component of agent test-time training.

本文提出了一种动作分级伤害评估量表，按七级有序量表对智能体的工具调用轨迹进行打分，评判依据包括所执行动作的可逆性、是否越界涉及其他方以及是否扩大了权限。An action-graded harm rubric is introduced that scores an agent's tool-call trajectory on a seven-level ordinal scale according to whether the executed action was reversible, whether it crossed scope to reach another party, and whether it expanded privilege.

本文在视觉语言导航、具身问答和语言条件操控任务上评估了三种 AAS 变体，覆盖四个具身执行器，结果表明架构级搜索能在具身任务上产生可部署且具有方向性的成功率提升，而其中一个看似得分较高的候选因存在泄漏而被判定为无效。This work evaluates three AAS variants across four embodied executors spanning vision-language navigation, embodied question answering, and language-conditioned manipulation, and shows that architecture-level search can produce deployable and directional success-rate gains on embodied tasks, while one apparent high-scoring candidate is rejected as leak-bearing.