[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-cybergym-glm-52-evidence-agent":3,"topics-all":38,"news-related-218e966c-1521-44eb-87d1-dba77cc26c9c":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"218e966c-1521-44eb-87d1-dba77cc26c9c","CyberGym 的 86.3%：GLM-5.2 安全 Agent 开始用证据说话","深信服AI在CyberGym全量1,507项真实漏洞任务中复现1,301项，成功率86.3%，并在官方复核后修正了3个误报。本文拆解固定使用GLM-5.2的Agent Swarm如何用证据状态、并行假设和对抗审查，把安全Agent从“会找线索”推进到“能提交可验证结论”，并讨论这比单纯模型跑分更值得关注的原因。","## CyberGym 的 86.3%：GLM-5.2 安全 Agent 开始用证据说话\n\n1,507 个真实漏洞，1,301 个被成功复现，成功率 86.3%。这个数字看起来像又一张大模型排行榜，但它真正有意思的地方，不在“GLM-5.2 跑了多少分”，而在于：一个安全 Agent 终于被放进了更接近真实工作的验证链里。\n\n### CyberGym 测的不是选择题\n\nCyberGym Level 1 收录了来自 188 个大型软件项目的 1,507 个历史漏洞，任务输入只有漏洞描述和补丁前的源代码。系统要做的不是回答“漏洞原因是什么”，而是分析代码、判断触发条件，最后构造一个 PoC，让它能在有漏洞的构建中触发，却不能在打过补丁的隐藏构建中触发。\n\n这一步很关键。只让程序崩溃，不能证明找对了漏洞；真正过关，必须同时满足“漏洞版本能复现、修复版本不能复现”。换句话说，评测把“看起来像对”与“经过独立验证确实对”分开了。\n\n### Sangfor AI 的核心，不只是换了一个模型\n\n这套系统固定使用 GLM-5.2，架构上采用 Agent Swarm：多个子 Agent 围绕不同的漏洞假设并行调查，但不把完整聊天历史粗暴拼进同一个上下文。它们共享的是证据状态，包括已经观察到的现象、被推翻的路径、触发条件和候选结论。\n\n这个设计解决的是安全自动化里一个经常被忽略的问题：多个 Agent 一致，不等于结论可靠。如果第一个 Agent 的猜测错了，后续 Agent 只是沿着同一条错路继续“表示赞同”，系统看起来很协同，实际上只是把错误放大。把假设与证据分开，保留负面结果，再让协调器根据证据缩小搜索空间，才有机会把并行探索变成真正的进展。\n\n接近提交时，系统还会引入对抗式候选审查，专门挑战 PoC 是否真的可复现、是否与被分配的漏洞对应。初步结果中有 1,304 个案例被确认，但 CyberGym 团队复核后认定其中 3 个是误报，最终数字修订为 1,301 个。这一处修订反而增强了结果可信度：系统没有把初始阳性全部包装成成绩，而是接受了外部验证带来的扣分。\n\n### 86.3% 的含义：安全 Agent 开始进入“证据时代”\n\n对企业来说，安全 Agent 的价值不是生成一份漏洞报告，而是减少人工排查仍要承担的“这到底是不是目标漏洞”的不确定性。对模型行业来说，这个案例也提醒我们，能力竞争正在从“谁的模型更聪明”转向“谁的系统更能被审计”。\n\n当然，86.3% 不能直接等同于现实生产环境的自动修复率。评测有明确的任务边界，运行环境、时间上限和工具权限也经过规定；而且每个任务最多提交一个最终 PoC。把 benchmark 成绩直接翻译成“企业可以无人值守地交给 Agent”，仍然过头了。\n\n但它确实说明了一件事：在高风险场景里，模型只是底座，证据治理才是系统能力。能提出十个假设不稀奇，能在提交前证明其中九个不该提交，才是安全 Agent 与普通聊天机器人的分水岭。\n\n这次值得记住的不是“GLM-5.2 赢了谁”，而是一个更硬的标准正在形成：AI 不该比谁更自信，而该比谁的结论更经得起隐藏环境的验证。","https:\u002F\u002Fwww.cybergym.io\u002Fcybergym\u002F","998df6db-96e6-4b8e-8be1-cfa00a6cd177",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"1fcfaaf2-67de-43d3-9e35-5784852fec60","ai-safety",{"id":19,"name":20,"slug":20,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"ddcdf4e9-85df-438d-93c3-ea9eb3c1b361","en","CyberGym's 86.3%: GLM-5.2's Security Agent Starts Speaking in Evidence","Sangfor AI reproduced 1,301 of 1,507 real-world vulnerability tasks in the full CyberGym evaluation, reaching an 86.3% success rate. The result is less important as a model leaderboard than as a demonstration of evidence-governed agent design: a fixed GLM-5.2 model, parallel vulnerability hypotheses, shared evidence states instead of shared chat history, adversarial candidate review, and hidden patched-build verification. After CyberGym reviewers removed three false positives from the initial count, the revised result became more credible. The case suggests that high-risk AI systems should compete on auditable, independently verified conclusions rather than confidence or raw benchmark scores.","## CyberGym’s 86.3%: GLM-5.2’s Security Agent Starts Speaking in Evidence\n\n1,507 real-world vulnerabilities, 1,301 successful reproductions, and an 86.3% success rate. At first glance, that looks like another entry on a large-model leaderboard. The more important story is different: GLM-5.2 was placed inside a security agent system whose conclusions had to survive an independent verification chain.\n\n### CyberGym is not a multiple-choice test\n\nCyberGym Level 1 contains 1,507 historical vulnerability tasks drawn from 188 large software projects. The system receives a vulnerability description and a pre-patch source repository. It must analyze the code, infer the actual trigger conditions, and construct a proof of concept. A submission passes only when the PoC triggers the vulnerable build but does not trigger the hidden patched build.\n\nThat distinction matters. Making a program crash is not enough to prove that the system found the assigned vulnerability. The hidden differential check separates “this looks plausible” from “this is confirmed to be the target defect.” It also makes the benchmark closer to a real security investigation, where an analyst must defend the connection between an observed failure and a specific root cause.\n\n### The important change is not just the model\n\nSangfor AI used GLM-5.2 as the fixed model throughout its Agent Swarm. Multiple agents investigated complementary vulnerability hypotheses in parallel, but the system did not simply concatenate every conversation into one oversized context. Instead, agents coordinated through an evidence state containing observations, rejected paths, trigger constraints, dependencies, and candidate claims.\n\nThis addresses a subtle failure mode in multi-agent systems: agreement does not equal reliability. If the first agent makes a wrong assumption, later agents can keep following the same path and appear highly collaborative while merely amplifying the original error. Separating hypotheses from evidence, retaining negative results, and allowing a coordinator to reduce the search space based on validated observations gives parallel exploration a chance to become real progress rather than synchronized speculation.\n\nAs a candidate approaches submission, the system also introduces adversarial review. The reviewer challenges whether the PoC is reproducible and whether it is actually tied to the assigned vulnerability. The initial evaluation confirmed 1,304 cases, but the CyberGym team later determined that three were false positives. The final result was revised to 1,301 reproductions. That correction makes the result stronger, not weaker: the system did not treat every initial positive as a success and accepted external verification as a reason to reduce the score.\n\n### What 86.3% really tells us\n\nFor an enterprise, the value of a security agent is not simply producing a vulnerability report. It is reducing the uncertainty that human analysts still have to resolve: is this the target defect, is the trigger reproducible, and will the proposed finding survive an independent check? For the model industry, the case points to a broader shift from model-centric capability claims toward system-level auditability.\n\nThe number should not be overinterpreted. An 86.3% benchmark score is not the same as an 86.3% autonomous remediation rate in production. CyberGym defines a particular task boundary, execution environment, time limit, tool policy, and submission protocol. Each task allows at most one designated final PoC. Real repositories are messier, asset inventories are incomplete, and production incidents rarely arrive with a clean vulnerability description.\n\nStill, the result establishes a useful direction. In high-risk domains, the model is only the substrate. Evidence governance is part of the capability. Generating ten plausible hypotheses is not the hard part; demonstrating that nine should not be submitted may be more valuable. That is the line separating a security agent from a chatbot that merely sounds certain.\n\nThe headline to remember is not that GLM-5.2 defeated a particular rival. It is that a harder evaluation standard is taking shape: AI systems should not compete on who sounds most confident. They should compete on whose conclusions can survive hidden-environment verification.","cybergym-glm-52-evidence-agent","2026-07-29T06:33:39Z","2026-07-29T06:48:21.235800Z","2026-07-29T06:48:21.235816Z",true,"agent",234,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"89de53a6-5cee-40af-8190-1c22d628b738","23 个端侧 LLM 同台比:Artificial Analysis 把 iPhone 17 Pro 变成首个开放基准","artificial-analysis-pipette-mobile-llm-benchmark","2026-09-01T11:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"47f12a0d-8559-473f-97be-dc12966bd4ff","DeepMind 双盲评测：Gemini 权重和考题锁进同一个加密飞地","deepmind-gemini-double-blind-eval","2026-08-29T15:05:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"b05de01b-89ca-499b-b130-e55162e651f5","SCOPE：让大模型学会选择性信任，而不是把上下文一概拒绝","scope-selective-trust-context-dpo","2026-08-06T17:59:58+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"86565410-ced9-4e64-80b7-d97a353a1d1d","LLM 元认知首次可度量：Metacognition-Bench 用 300 道陷阱题 + 11 个开源错题雷达适配器把自我纠错做成开源工程","metacognition-bench-llm-self-correction","2026-07-01T12:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"817a35b2-6b31-41e6-a213-3ac6f667fd14","MosaicLeaks：ServiceNow 撕开 Deep Research Agent 的\"查询即泄密\"盲区","mosaicleaks-servicenow-deep-research-leak","2026-06-26T16:30:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"43a0972c-f040-483c-88cd-89b40a22723d","RA-RFT：检索增强强化微调，让 LLM 学会「类比推理」而非「按字找参考」","ra-rft-retrieval-augmented-analogical-rl","2026-06-12T18:30:00+00:00"]