Daily Papers · 2026-09-06
📄 论文日报 | 来源：Hugging Face Daily Papers
🗓 Episode：【月末特辑】8月最火AI论文 | 循环潜在推理兼顾上下文学习；执行框架扩展提升智能体表现
1. BDH-CQ：基于循环潜在推理的上下文学习
- 英文标题：BDH-CQ: In-Context Learning with Recurrent Latent Reasoning
- arXiv：2608.09888
💡 一句话解读：这篇研究要解决的是如何让AI在推理时更高效地利用示例来解决问题。他们提出了一种新方法，让AI在内部反复思考，而不是把思考过程说出来，从而在成本很低的情况下提高了在ARC-AGI-1测试上的准确率，刷新了性价比纪录。
我们提出了BDH-CQ，一种结合上下文学习（in-context learning）与循环潜在推理（recurrent latent reasoning）的推理模型。推理时输入的示例会持续更新模型的循环记忆；随后，模型通过在高维潜在空间中的迭代计算来解答查询，而无需将中间推理过程言语化。我们在公开的ARC-AGI-1评估集上评估了该模型，并使用受控的类ARC干预来研究它从演示中学到了什么、如何一致地应用推断出的变换，以及哪些概念仍然难以处理。一个150M参数的配置在每任务计算成本为0.0007美元的情况下达到了29.5%的pass@2，这一操作点突破了先前报道的ARC-AGI-1成本-准确率帕累托前沿，确立了基准成本效率的新最优水平。
📄 English Abstract
We introduce BDH-CQ, a reasoning model that combines in-context learning with recurrent latent reasoning. Inputs presented at inference time continuously update the model’s recurrent memory; the model then solves a query through iterative computation in a high-dimensional latent space, without verbalizing its intermediate reasoning. We evaluate the model on the public ARC-AGI-1 evaluation set and use controlled ARC-like interventions to study what it learns from demonstrations, how consistently it applies an inferred transformation, and which concepts remain difficult. A 150M-parameter configuration reaches 29.5% pass@2 at a computed inference cost of $0.0007 per task. This operating point breaks through the previously reported ARC-AGI-1 cost-accuracy Pareto frontier, establishing a new state of the art in benchmark cost efficiency.
2. StateM：通过 Harness 扩展在 Terminal-Bench 2.1 上达到 95.3% 原始准确率，或一次 15 美元的前沿运行
- 英文标题：StateM: Reaching 95.3% Raw Accuracy, or a $15 Frontier Run, on Terminal-Bench 2.1 via Harness Scaling
- arXiv：2608.15089
💡 一句话解读：这篇研究要解决的是：AI 智能体在执行长任务时，即使模型本身很聪明，也常会忘记步骤、中途出错或提前停止。他们提出的方法是：不改变模型，而是改进智能体周围的“执行系统”，像给任务加一个“操作手册”和“状态追踪器”，让每一步都有记录、可检查、能恢复。结果在测试中大幅提升了准确率，成本还很低。
长程智能体（long-horizon agents）即使其底层模型能够解决各个组成步骤，也可能失败。它们可能会丢失可变状态（mutable state）的跟踪，未能重新激活早期执行中的经验，跳过已知程序，或过早停止。我们押注于通过扩展执行系统（harness scaling）来改进智能体周围的执行环境，而不改变其模型权重。我们引入了 StateM，一个智能体原生运行时（agent-native runtime），它围绕持久状态（durable states）、阶段局部上下文（phase-local context）、检查过的转换（checked transitions）、可恢复运行手册（recoverable runbooks）以及智能体和用户可共同检查的版本化程序实践（versioned procedural practices）来组织执行。在 Terminal-Bench 2.1 上，StateM 将 GPT-5.5 xhigh 提升至 92.1%，而参考值为 83.1%，GPT-5.6 Sol Ultra 为 91.9%。该运行手册无需修改即可迁移至 GPT-5.6。使用 GPT-5.6 Sol xhigh，StateM 在 445 次试验中达到 95.3% 的原始准确率，并在所有 89 个任务中至少成功一次。冻结的配置文件将 GPT-5.6 Luna 从 76.7% 提升至 85.4%，高于 Sol xhigh 的 84.9% 参考值。使用相同的运行时、运行手册结构和黄金规则，不到 38 美元的适配成本将 DeepSeek-V4 Flash 在标准超时下从 82.7% 提升至 88.1%，在 88 个任务的公共核心上提升至 89.1%。仅扩展剩余的延迟敏感任务即可匹配报告的 88.8% GPT-5.6 Sol 最大结果。最终评分的 API 使用费用约为 15 美元，而 GPT 参考为 574.68 美元；DeepSeek 总支出为 52.22 美元。在 BusinessBench 上，基于开发集构建的特定家族运行手册在保留集上获得了 0.55 个宏平均点和 1.34 个微平均点的提升；两个机制匹配的家族提升了 10.04 个点。当任务共享执行结构时，具体规则能够泛化，而控制方法则广泛适用。StateM 将选定的事后分析发现转化为持久、可执行的先决条件和实践，通过有状态控制使学习到的控制变得明确且可强制执行。代码见 github.com/henryqin1997/statem。
📄 English Abstract
Long-horizon agents can fail even when their underlying models can solve the constituent steps. They may lose track of mutable state, fail to reactivate lessons from earlier executions, skip known procedures, or stop prematurely. We bet on harness scaling to improve the execution system around an agent without changing its model weights. We introduce StateM, an agent-native runtime that organizes execution around durable states, phase-local context, checked transitions, recoverable runbooks, and versioned procedural practices that agents and users can inspect together. On Terminal-Bench 2.1, StateM raises GPT-5.5 xhigh to 92.1%, versus 83.1% reference and GPT-5.6 Sol Ultra at 91.9%. The runbook transfers unchanged to GPT-5.6. With GPT-5.6 Sol xhigh, StateM reaches 95.3% raw accuracy across 445 trials and succeeds on all 89 tasks at least once. The frozen profile raises GPT-5.6 Luna from 76.7 to 85.4%, above the 84.9% Sol xhigh reference. Using the same runtime, runbook structure, and golden rules, less than $38 of adaptation raises DeepSeek-V4 Flash from 82.7 to 88.1% under standard timeouts and to 89.1% on an 88-task common core. Extending only the remaining latency-sensitive task matches the reported 88.8% GPT-5.6 Sol max result. Final-score API usage is about $15 versus $574.68 for the GPT reference; total DeepSeek expenditure is $52.22. On BusinessBench, family-specific runbooks built on development sets yield held-out gains of 0.55 macro and 1.34 micro points; two mechanism-matched families improve by 10.04 points. Concrete rules generalize when tasks share execution structure, while the control methodology applies broadly. StateM turns selected postmortem findings into persistent, executable preconditions and practices, making learned controls explicit and enforceable through stateful controls. Code at github.com/henryqin1997/statem.
3. Macaron-V1：迈向具备自我改进和LoRA混合的开放持续学习
- 英文标题：Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA
- arXiv：2608.09819
💡 一句话解读：这篇论文介绍了一个名为Macaron-V1的AI模型家族，它能在实际环境中不断学习改进。主要解决AI部署后无法持续学习的问题。方法是采用冻结基础模型并组合多个专业模块（LoRA）的架构，以及一个自我改进的循环系统。
Macaron-V1是一个用于经验智能的开放智能体-模型家族：从真实环境中的经验学习，并在部署后继续学习。它围绕两个系统目标组织。适应性通过版本化模型-工具对（model-harness pairs）的递归改进来实现，其中一种配置的经验在外部契约下评估，并用于构建其继任者。协作通过LoRA混合（Mixture-of-LoRA, MoL）架构实现，该架构冻结基础模型，组合专家LoRA适配器，并为每个用户轮次选择一个LoRA。旗舰版Macaron-V1-Venti（748B）将744B的GLM-5.2基础模型与四个用于聊天、智能体、编码和GenUI的LoRA相结合；基于Qwen3.6-35B的Macaron-V1-Tall（50B）采用相同设计用于本地部署。本报告将Macaron-V1呈现为一个涵盖架构、算法和基础设施的协同设计系统。MoL架构通过可扩展的LoRA专家支持持续学习。该算法结合了模型-工具协同设计（Model-Harness Co-design）和递归自我改进循环，包括UI4A组件原生GenUI工具、有状态动作基底、版本化工具上下文协议（Harness Context Protocol）契约，以及智能体强化学习框架MindForge。支持性基础设施包括后训练平台MinT、长上下文强化学习方法LongStraw，以及用于稀疏MoE和DSA基础模型的稳定性技术。我们在个人智能、GenUI和通用能力基准上对Macaron-V1进行了评估，并与前沿基线进行了比较。我们的结果验证了当前系统，而持续学习和集体智能带来的复合收益仍是开放问题。
📄 English Abstract
Macaron-V1 is an open agent-model family for experiential intelligence: learning from experience in real environments and continuing to learn after deployment. It is organized around two system goals. Adaptation is pursued through recursive improvement of versioned model-harness pairs, where experience from one configuration is evaluated under an external contract and used to construct its successor. Collaboration is pursued via the Mixture-of-LoRA (MoL) architecture that freezes a base model, composes specialist LoRA adapters, and selects one LoRA per user turn. The flagship Macaron-V1-Venti (748B) combines a 744B GLM-5.2 base with four LoRAs for chat, agent, coding, and GenUI; the Qwen3.6-35B-based Macaron-V1-Tall (50B) uses the same design for local deployment. This report presents Macaron-V1 as a co-designed system spanning architecture, algorithms, and infrastructure. The MoL architecture supports continual learning through extensible LoRA specialists. The algorithm combines Model-Harness Co-design and recursive self-improvement loop, including the UI4A component-native GenUI harness, a stateful action substrate, versioned Harness Context Protocol contract, and the agentic RL framework MindForge. The supporting infrastructure includes the post-training platform MinT, the long-context RL method LongStraw, and stability techniques for sparse MoE and DSA base models. We evaluate Macaron-V1 on Personal Intelligence, GenUI, and general capability benchmarks against frontier baselines. Our results validate the current system, while compounding gains from continual learning and collective intelligence remain open questions.
4. HarnessEval-W：使视觉世界的评估智能体化
- 英文标题：HarnessEval-W: Agentifying the Evaluation of Visual Worlds
- arXiv：2608.16859
💡 一句话解读：这篇研究要解决的是：现有世界模型评估只给一个分数，却说不清为什么，让人难以信任。它提出的方法是：像让一群专家分工检查一样，把评估拆成小问题，由专门的小助手分别验证，最后汇总成有依据的结论。这样每次评估都留下清晰的推理过程，结果更可信。
一个基准测试应提供的不只是一个分数：使评估可信的是证明该分数的推理过程。这对于世界模型（world models）尤为关键，因为判断一次生成需要理解物理、因果关系和世界状态是否正确演化。人类能自然发现此类违规，但现有基准测试无法自动化这一能力：指标通过暴力计算得出，没有可检查或验证的推理链。我们引入了HarnessEval-W，一个智能体化的评估流程，将LLM生态系统中的“驾驭”（harness）范式带到世界模型基准测试中。HarnessEval-W不采用固定规则，而是解释每个评估案例的上下文，将评估问题分解为可测量的子问题，并生成专门的子智能体，每个子智能体配备定制上下文和诊断工具来推理其子问题。父智能体随后验证收集的证据并将其总结为最终裁决。这一分层工作流将每次评估转化为透明的证据树，其完整推理链为结果提供依据。我们将HarnessEval-W应用于18个代表性世界模型，涵盖330个评估案例。其判断与人类偏好高度一致，同时为每次生成提供可验证的细粒度诊断。我们开源了整个流程作为实时基准测试，并邀请广泛社区随着世界模型的演进贡献新技能和评估案例。
📄 English Abstract
A benchmark should deliver more than a scalar score: what makes an evaluation trustworthy is the reasoning that justifies the score. This is especially critical for world models, where judging a rollout requires understanding whether physics, causality, and world state evolve correctly. Humans spot such violations naturally, yet no existing benchmark automates this capability: metrics are computed brute-force, leaving no reasoning chain that can be examined or verified. We introduce HarnessEval-W, an agentified evaluation pipeline that brings the harness paradigm from the LLM ecosystem to world model benchmarking. Rather than applying a fixed rubric, HarnessEval-W interprets the context of each evaluation case, decomposes the evaluation question into measurable subproblems, and spawns specialized sub-agents, each equipped with tailored context and diagnostic tools to reason over its own subproblem. The parent agent then validates the gathered evidence and summarizes it into the final verdict. This hierarchical workflow turns every evaluation into a transparent evidence tree whose complete reasoning chain justifies the result. We apply HarnessEval-W to 18 representative world models over 330 evaluation cases. Its judgments closely align with human preferences while providing verifiable, fine-grained diagnoses of every generated rollout. We open-source the full pipeline as a live benchmark and invite the broad community to contribute to grow new skills and evaluation cases as world models evolve.
5. 从火花到论文：作为可组合技能的端到端研究论文生成
- 英文标题：Spark-to-Paper: End-to-End Research Paper Generation as a Composable Skill
- arXiv：2608.11924
💡 一句话解读：这篇研究要解决如何让AI自动写完整科研论文的问题，不只是写文字，还要查文献、做实验、画图并保证前后一致。他们提出了一个叫Spark-to-Paper的系统，把它做成13个可组合的步骤，集成在现有编程助手里，不需要额外平台。系统把AI判断和可检查的确定性操作分开，并用检查机制防止实验反复推翻原目标。结果在8个主题上，引用准确率99.5%，图表可编辑率96.4%，成本约8美元，耗时3.2小时。
将研究想法转化为完整论文需要的不仅仅是文本生成：系统必须检索文献、设计和执行实验、根据证据修改论断、生成可发表的图表，并在长生成过程中保持一致性。我们提出了Spark-to-Paper，一个端到端的研究论文生成系统，在现有编码助手中实现为十三个可组合技能，无需单独的代理平台或编排服务。Spark-to-Paper将基于模型的判断与可直接执行和检查的确定性操作分开。它还将实验规划与报告分开，以便在观察结果之前指定所需证据，并根据测量结果修改手稿论断。为提高长研究轨迹的可靠性，系统将确定性完整性检查与自我批评相结合，并限制了一种我们称之为自我反驳循环的失败模式，在该模式中，重复实验持续拒绝原始研究目标。Spark-to-Paper还通过程序化绘图生成实验结果的可编辑矢量图，并通过基于代码的重建生成方法示意图。在八个受控研究主题中，Spark-to-Paper实现了99.5%的引用有效性和96.4%的图表可编辑性。一项受控消融研究将伪造检测率从单遍草稿的14%提高到完整完整性和审查堆栈的92%，而对抗性审查实现了74%的精确度。完整系统使用1190万token，每篇手稿成本8.1美元，平均需要3.2小时。这些结果表明，端到端研究论文生成可以在现有编码助手中实现为轻量级、可组合的工作流，同时保持实验证据对论断接受、修改或放弃的核心作用。
📄 English Abstract
Turning a research idea into a complete paper requires more than text generation: the system must retrieve literature, design and execute experiments, revise claims according to evidence, produce publication-ready figures, and maintain consistency across a long generation process. We present Spark-to-Paper, an end-to-end research paper generation system implemented as thirteen composable skills inside an existing coding assistant, without requiring a separate agent platform or orchestration service. Spark-to-Paper separates model-based judgment from deterministic operations that can be directly executed and checked. It further separates experiment planning from reporting, so that required evidence is specified before results are observed and manuscript claims are revised according to measured outcomes. To improve reliability over long research trajectories, the system combines deterministic integrity checks with self-critique and bounds a failure mode we call the Self-Refutation Loop, in which repeated experiments continue to reject the original research objective. Spark-to-Paper also produces editable vector figures through programmatic plotting for experimental results and code-based reconstruction for generated method diagrams. Across eight controlled research topics, Spark-to-Paper achieves 99.5% citation validity and 96.4% figure editability. A controlled ablation increases fabrication detection from 14% for a single-pass draft to 92% with the full integrity and review stack, while adversarial review achieves 74% precision. The full system uses 11.9M tokens, costs $8.1 per manuscript, and requires 3.2 hours on average. These results show that end-to-end research paper generation can be implemented as a lightweight, composable workflow inside existing coding assistants while keeping experimental evidence central to how claims are accepted, revised, or abandoned.
6. 我们能防御针对现实世界危机事件的AI生成视频攻击吗？对检测器、生成器与社会传播的系统评估
- 英文标题：Can We Defend Against AI-Generated Video Attacks on Real-World Crisis Events? A Systematic Evaluation of Detectors, Generators and Social Dissemination
- arXiv：2608.14391
💡 一句话解读：这篇研究要解决的是：AI生成的假视频在战争、灾难等真实危机事件中可能误导公众，而现有检测工具不够可靠。他们创建了一个大型测试集RA-Bench，包含大量真实和AI生成的视频，并系统测试了多种检测器、生成条件和人类判断。结果发现，现有检测器很难稳定识别这些逼真假视频，社交媒体传播更让检测难上加难。
最近的视频生成器可以制造出关于战争、灾难、公共紧急事件等现实世界危机的逼真描绘，带来重大的错误信息风险。然而，现有基准在检测器和生成器在此类情境下的行为方面提供的证据有限，包括可检测性如何随生成条件变化、人们如何感知生成的视频，以及检测器在社交传播过程中是否仍然可靠。为弥补这一空白，我们引入了RA-Bench，一个以真实视频为锚点的AI生成视频检测基准。RA-Bench包含17,886个视频，涵盖10个社会风险类别中的1,830个真实视频锚点，以及来自四个开源和五个闭源生成器的16,056个生成片段。基于RA-Bench，我们从三个维度组织评估。首先，我们评估了七个传统检测器、三种审查设置下的十个零样本多模态模型，以及两个专门针对AI生成视频检测进行微调的多模态大语言模型（MLLM）的检测泛化能力。在这些方法中，三个检测器家族均未能在RA-Bench实例上一致地泛化。接着，我们考察可检测性如何随生成质量、条件信息和采样种子变化。这些分析表明，生成属性对不同检测器家族的影响不同，而源级检测模式在不同种子间保持稳定。最后，我们研究了社交传播期间的人类真实性判断和检测器可靠性。我们发现，误导人们的视频对当前检测器来说也难以识别，且社交传播使检测更加困难。总之，这些发现表明，当前方法难以检测逼真的AI生成视频，凸显了对能够适应不断演变的视频生成器的稳健检测器的需求。
📄 English Abstract
Recent video generators can fabricate realistic depictions of wars, disasters, public emergencies, and other real-world crises, creating substantial risks of misinformation. Existing benchmarks, however, provide limited evidence on detector and generator behavior in such settings, including how detectability varies with generation conditions, how people perceive generated videos, and whether detectors remain reliable during social dissemination. To address this gap, we introduce RA-Bench, a benchmark for AI-generated video detection that uses Real videos as Anchors. RA-Bench contains 17,886 videos, comprising 1,830 real-video anchors across 10 social-risk categories and 16,056 generated clips from four open-source and five closed-source generators. Based on RA-Bench, we organize our evaluation along three dimensions. We first assess detector generalization across seven traditional detectors, ten zero-shot multimodal models under three review settings, and two MLLMs specifically fine-tuned on AI-generated video detection. Across these methods, none of the three detector families generalizes consistently across RA-Bench instances. We then examine how detectability varies with generation quality, conditioning information, and sampling seeds. These analyses show that generation properties affect detector families differently, while source-level detection patterns remain stable across seeds. Finally, we study human authenticity judgments and detector reliability during social dissemination. We find that videos that mislead people are also difficult for current detectors, and that social dissemination makes detection harder. Together, these findings show that current methods struggle to detect realistic AI-generated videos, highlighting the need for detectors robust to evolving video generators.
7. EnvHarness：为智能体学习唤醒静态世界
- 英文标题：EnvHarness: Awakening Static Worlds for Agent Learning
- arXiv：2608.19880
💡 一句话解读：这篇研究要解决的是：AI智能体在训练时使用的环境是固定不变的，无法根据智能体的弱点动态调整，导致训练效率低。他们提出了一种叫EnvHarness的方法，像给环境加装可插拔的“插件”，在不改动环境核心逻辑的情况下，动态调整环境难度和内容，让智能体持续进步。
LLM智能体通过与环境的交互来学习，但这些环境是手工构建且静态的，对智能体的弱点视而不见，并随着智能体的进步而迅速落后。虽然最近的环境生成方法试图解决这个问题，但它们需要特定领域的流程，依赖昂贵或不可靠的验证器，并且仍然产生静态环境。为了减轻从头重建环境的工程负担，我们提出了环境束（EnvHarness），一个可编程的插件组件层，它包裹静态环境以重塑其行为，而无需修改底层逻辑。通过标准接口操作，EnvHarness适用于不同领域，同时确保每个重塑后的环境保留其原始验证器。为了自动化这一过程，我们引入了EnvRigger，它将目标策略视为黑盒，观察其执行轨迹以合成针对诊断缺陷的EnvHarness组件，并通过新的rollout进行验证。在四个领域的五个基准测试中，EnvHarness优于原始环境和特定领域的环境生成流程，在保留实例上实现了高达9.0分的提升，同时执行步骤减少了9.8%。此外，EnvHarness为强化学习提供了更优的优化信号，使策略及其环境能够持续、有针对性地共同进化。
📄 English Abstract
LLM agents learn by interacting with environments, yet these environments are hand-built and static: blind to an agent’s weaknesses, and quickly left behind as it improves. While recent environment generation methods attempt to address this, they require domain-specific pipelines, rely on expensive or unreliable verifiers, and still produce static environments. To alleviate the engineering burden of rebuilding environments from scratch, we propose Environment Harness (EnvHarness), a programmable layer of plug-in components that wraps a static environment to reshape its behavior without modifying the underlying logic. Operating through standard interfaces, EnvHarness applies across diverse domains while ensuring every reshaped environment retains its original verifier. To automate this process, we introduce EnvRigger, which treats the target policy as a black box, observing its execution trajectories to synthesize EnvHarness components targeting diagnosed flaws, and validating them via fresh rollouts. Across five benchmarks in four domains, EnvHarness outperforms both original environments and domain-specific environment generation pipelines, achieving up to a 9.0-point improvement on held-out instances with 9.8% fewer execution steps. Furthermore, EnvHarness provides a superior optimization signal for reinforcement learning, enabling continuous, targeted co-evolution of the policy and its environment.
8. VBVR-Pro：可扩展且可验证的原生视觉推理套件
- 英文标题：VBVR-Pro: A Scalable and Verifiable Suite for Native Visual Reasoning
- arXiv：2608.26105
💡 一句话解读：这篇研究要解决的是：如何让AI通过生成图像和视频来进行推理，而不仅仅是理解它们。他们提出了一个叫VBVR-Pro的测试平台，包含300个程序化生成的任务，并提供可靠的评分系统来验证AI的推理结果。实验表明，在这个平台上训练的AI模型在其他视觉推理测试中表现更好，并且视频生成在需要跟踪时空状态的任务中效果最佳。
原生视觉推理将视觉生成视为推理本身的媒介：视觉状态（即图像和视频）不仅是需要理解或渲染的输入输出，而且是超越语言的解决问题的第一类基质。然而，由于缺乏可扩展的训练任务、可靠的反馈以及跨生成基质的受控比较，进展仍然受阻。在这项工作中，我们引入了VBVR-Pro，一个闭环测试平台，使通过生成进行的原生视觉推理变得可训练、可验证、可优化和实验可控。1) 任务扩展：VBVR-Pro将视觉推理转化为一个包含300个程序化生成任务的受控任务空间。在VBVR-Pro上训练的模型在七个外部视觉推理基准（如RISE-Video、MME-CoF-Pro和BabyVision）上显示出超越所提套件的强迁移能力。2) 可验证奖励：VBVR-Pro为任务基础评估提供了可验证的奖励评分器。通过对领先的MLLM作为评判者的系统研究，我们识别出流行的VLM-as-a-judge范式的常见失败模式。相比之下，所提出的评分器基于确定性的、任务特定的规则，与人类判断实现了细粒度对齐。重要的是，它们作为大规模多任务强化学习的可靠奖励信号，并在视觉推理任务中展示了更强的后强化学习性能。3) 机制研究：VBVR-Pro支持对超过30种图像、视频和交错生成器的受控模态研究。我们的分析表明，视频生成在需要持久时空状态跟踪的任务中仍然最强，而交错生成提供了一种计算高效的替代方案。关键的是，消融和探测表明存在对视觉推理至关重要的视觉原生轨迹。我们发布了所有数据、模型、评分器和代码。
📄 English Abstract
Native visual reasoning treats visual generation as the medium of reasoning itself: visual states (i.e. images and videos) are not merely inputs to be understood or outputs to be rendered, but first-class substrates for problem solving beyond language. Yet progress remains bottlenecked by the lack of scalable training tasks, reliable feedback, and controlled comparisons across generative substrates. In this work, we introduce VBVR-Pro, a closed-loop testbed that makes native visual reasoning through generation trainable, verifiable, optimizable, and experimentally controllable. 1) Task scaling. VBVR-Pro turns visual reasoning into a controlled task space of 300 procedurally generated tasks. Models trained on VBVR-Pro show strong transfer beyond the proposed suite across seven external visual reasoning benchmarks such as RISE-Video, MME-CoF-Pro, and BabyVision. 2) Verifiable rewards. VBVR-Pro provides verifiable reward scorers for task-grounded evaluation. Through a systematic study of leading MLLMs as judges, we identify recurring failure modes of the prevalent VLM-as-a-judge paradigm. In contrast, the proposed scorers are grounded in deterministic, task-specific rules, achieve fine-grained alignment with human judgments. Importantly, they serve as reliable reward signals for large-scale multi-task reinforcement learning and demonstrate stronger post-RL performance across visual reasoning tasks. 3) Mechanism study. VBVR-Pro enables controlled modality studies across more than 30 image, video, and interleaved generators. Our analysis shows that video generation remains strongest for tasks requiring persistent spatiotemporal state tracking, while interleaved generation provides a compute-efficient alternative. Critically, ablations and probing suggest the presence of vision-native trajectories that are crucial to visual reasoning. We release all data, models, scorers, and code.
9. OpenART：通过开放式环境演化扩展智能体红队测试
- 英文标题：OpenART: Scaling Agent Red Teaming via Open-Ended Environment Evolution
- arXiv：2608.00677
💡 一句话解读：这篇研究要解决的是：现有AI安全测试只针对简单、静态的任务，无法发现AI智能体在长期、动态环境中累积的风险。他们提出了一个叫OpenART的测试平台，包含超过1万个复杂场景，并设计了一种叫EMHA的攻击方法，能自动演化环境来更有效地暴露AI的安全漏洞。
AI智能体在持久环境中运行，早期的状态变化可能影响未来的决策。与传统语言模型交互不同，智能体的行为通过共享状态进行中介，该状态在长时间跨度的工作流程中被反复修改和重用。当前的安全基准往往无法捕捉这些累积风险，因为它们专注于短期的静态任务。为解决这些局限性，我们引入了OpenART，一个通过环境演化进行可扩展智能体红队测试的开放式竞技场。OpenART提供了跨越50个领域的超过10,000个经过验证的有状态场景，来源于超过500,000个工具和技能的池子。这些任务需要中位数97次工具调用，并支持在75种不同的智能体模型配置上进行统一评估。为了系统探索这些不断演化的攻击面，我们提出了演化马尔可夫超图攻击（EMHA）。EMHA是一种黑盒策略，通过协调授权的状态转换进行反馈驱动的环境演化，无需参数更新。在整个评估过程中，任务目标保持不变，只有环境状态发生变化。在所有配置中，EMHA实现了85.0%的汇总攻击成功率（ASR）。其相对于仅指令演化的优势从简单环境中的约2%增加到最复杂环境中的超过17%，表明随着任务复杂性的增长，环境演化越来越能暴露安全失败。此外，我们的分析显示，智能体的具体运行时实现解释了超出底层模型能力之外的安全差异的很大一部分。这些结果确立了OpenART作为在复杂、演化环境中研究智能体安全的可扩展基础。
📄 English Abstract
AI agents operate in persistent environments where early state changes can influence decisions far into the future. Unlike conventional language-model interactions, agent behavior is mediated through a shared state that is repeatedly modified and reused across long-horizon workflows. Current safety benchmarks often fail to capture these cumulative risks because they focus on short, static tasks. To address these limitations, we introduce OpenART, an open-ended arena for scalable agent red teaming through environment evolution. OpenART provides over 10,000 validated stateful scenarios across 50 domains, drawing from a pool of more than 500,000 tools and skills. These tasks require a median of 97 tool calls and enable unified evaluation across 75 different agent-model configurations. To systematically explore these evolving attack surfaces, we propose the Evolutionary Markov Hypergraph Attack (EMHA). EMHA is a black-box policy that performs feedback-driven environment evolution by coordinating authorized state transitions without requiring parameter updates. Throughout the evaluation, task objectives remain fixed while only the environment state changes. Across all configurations, EMHA achieves a pooled Attack Success Rate (ASR) of 85.0%. Its advantage over instruction-only evolution increases from approximately 2% on simple environments to over 17% on the most complex ones, demonstrating that environment evolution increasingly exposes safety failures as task complexity grows. Furthermore, our analysis shows that the specific runtime implementation of an agent explains a significant portion of safety variation beyond the underlying model’s capabilities. These results establish OpenART as a scalable foundation for studying agent safety in complex, evolving environments.
10. 面向长时程终端任务的递归式合成
- 英文标题：Recursive Synthesis for Long-Horizon Terminal Tasks
- arXiv：2608.05466
💡 一句话解读：这篇研究要解决的是：为终端AI代理生成高质量长时程训练数据太贵太难。他们提出一种递归合成方法，从已验证的小任务开始，逐步扩展成更复杂的大任务，并自动检查正确性。这样能以极低成本生成大量高难度任务，用于训练AI，显著提升其性能。
为终端代理生成高质量的长时程训练数据成本高昂，每个任务需花费数百至数千美元，因为必须保持指令、环境、参考解决方案和验证器相互一致。人工编写无法扩展，而直接使用大型语言模型（LLM）生成往往破坏这些依赖。我们提出了递归合成终端任务（RST），一种递归验证的合成框架，用于大规模构建长时程终端代理任务。从已验证的种子任务开始，RST扩展参考解决方案，将验证器和指令与新工作流重新对齐，在新沙箱中验证结果，并将接受的任务作为后续轮次的种子。经过十五轮递归，RST以约每任务0.05美元的成本生成了37,484个合成终端代理任务。任务难度随轮次显著增加：参考解决方案的中位数从67行增至374行，执行命令的中位数从40增至244，DeepSeek-V4-Pro的pass@4从R1的90%降至R15的2.5%。为展示训练效用，我们在合成任务上收集了拒绝采样的Qwen3.5轨迹，并用于监督微调。在这些轨迹上微调使Qwen3.5-27B和Qwen3.5-122B-A10B在Terminal-Bench 2、Terminal-Bench Hard和Long-Horizon Terminal Bench上提升高达10分，而基于代理的PPO将Qwen3.5-27B在三个基准上分别提升至49.44%、32.00%和22.07%，相对基础模型分别提升20.0%、41.2%和21.9%。此外，15轮后递归未显示上限：合成产量和验证率保持稳定，而难度持续攀升，表明该过程可远超此处报告的规模继续运行。
📄 English Abstract
High-quality long-horizon training data for terminal agents is expensive to produce, often costing hundreds to thousands of dollars per task, because each task must keep the instruction, environment, reference solution, and verifier mutually consistent. Human authoring does not scale, and direct generation with large language models (LLMs) often breaks these dependencies. We present Recursive Synthetic Terminal Tasks (RST), a recursive verified synthesis framework for constructing long-horizon terminal-agent tasks at scale. Starting from verified seed tasks, RST extends the reference solution, realigns the verifier and instruction to the new workflow, validates the result in a fresh sandbox, and reuses accepted tasks as seeds for subsequent rounds. Across fifteen recursive rounds, RST produces 37,484 synthesized terminal-agent tasks at roughly $0.05 per task. Task difficulty increases substantially over rounds: the median reference solution grows from 67 to 374 lines, the median number of executed commands grows from 40 to 244, and DeepSeek-V4-Pro pass@4 drops from 90% at R1 to 2.5% at R15. To demonstrate training utility, we collect rejection-sampled Qwen3.5 trajectories on the synthesized tasks and use them for supervised fine-tuning. Fine-tuning on these trajectories improves Qwen3.5-27B and Qwen3.5-122B-A10B by up to 10 points on Terminal-Bench 2, Terminal-Bench Hard, and Long-Horizon Terminal Bench, while agentic PPO lifts Qwen3.5-27B to 49.44%, 32.00%, and 22.07% on the three benchmarks, corresponding to relative gains of 20.0%, 41.2%, and 21.9% over the base model. Moreover, after 15 rounds, the recursion shows no ceiling: synthesis yield and validation rates remain stable as difficulty keeps climbing, indicating that the process can continue well beyond the scale reported here.