[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-google-freezes-oss-vrp-ai-bug-slop":3,"topics-all":35,"news-related-db858586-4427-4480-bfe0-fa7747ab6d10":54},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":21,"news_slug":28,"published_at":29,"created_at":30,"modified_at":31,"is_published":32,"publish_type":33,"image_url":14,"view_count":34},"db858586-4427-4480-bfe0-fa7747ab6d10","Google 冻结 OSS VRP：AI 报告淹没开源赏金","2026 年 10 月 1 日,Google 旗下 OSS VRP 的产品漏洞提交入口被暂停,理由是大量 AI 生成的报告涌入且绝大多数无效,工程师不堪重负。arXiv、Internet Bug Bounty 与 Linux 内核都正遭受同类冲击,开源安全协作机制被 AI slop 反噬。","2026 年 10 月 1 日,Google 旗下针对自家开源项目的漏洞赏金计划 OSS VRP 正式冻结「产品漏洞」提交入口,给出的理由非常直白:涌入的 AI 生成报告太多,绝大多数无效,处理它们本身成了负担。这是全行业被「AI slop」反噬的又一个典型场景——去年还在帮安全研究员提效的工具,今年开始把研究员本身淹没。\n\n## OSS VRP 暂停的是什么、保留的是什么\n\n按 Google 的官方说明,暂停范围明确限定在「产品漏洞」一类,10 月 1 日之前已经提交的报告不受影响,供应链攻击报告继续接收,Cloud VRP 与 Patch Rewards Program 仍然开放。换句话说,Google 没有停掉整个开源安全协作,只是把最容易被低质 AI 报告灌水的入口暂停,承诺在 2027 年一季度给出调整方案。从规则页直接拿掉的「产品漏洞」奖励等级看,OT0 到 OT3 四个项目层级目前都没有金额显示,等于是把这个口子直接关闭。\n\n## 为什么 AI 报告变得不可承受\n\nOSS VRP 覆盖的项目包括 Go、Angular、Protocol Buffers 等核心开源仓库,过去的研究员要做出能被采纳的提交,需要在代码层真正定位缺陷、设计错误或供应链组件漏洞,这是需要技巧的苦活。而大模型和自动化挖洞脚本的普及,把提交的「成本-技巧」比几乎压扁——任何人都可以让 LLM 扫一遍代码、产出一份格式上看似合理的报告。Help Net Security 在 5 月就对这类问题做过专题,Google 这次暂停实际上是对当时已蔓延数月的工程师抱怨的官方回应。\n\n维护者们要花大量时间人工甄别报告,判断「这是个真漏洞」还是「模型在不熟悉的边界情况上产生的幻觉」。后果是,真正严重的漏洞反而被排队等处理,开源安全协作的核心信任机制被消耗。\n\n## 不只是 Google 一家\n\n行业里被同一波冲击的还有另外两家值得注意:\n\n- **arXiv** 在同一天 10 月 1 日开始执行「每月两篇提交上限」的措施,理由是「高密度、近乎自动化的投稿大量涌入」。\n- **Internet Bug Bounty (IBB)** 由 HackerOne 代管的公共漏洞池早在今年 3 月 27 日就暂停了新提交,该计划覆盖 curl 等核心项目的赏金发放,在 HackerOne API 里截至 10 月 6 日仍标为 paused。\n- **Linux 内核**本月公开表示维护者「完全不堪重负」,在新版本里一次性推进到的 CVE 数量达到创纪录的 2000 个,主要是 AI 辅助挖洞脚本的产物。\n- **Intel** 早些时候暂停了单个漏洞最高付 10 万美元的漏洞赏金,未官方确认 AI 报告是原因,但行业普遍持此推测。\n\n这意味着「AI 改造科研与工程流程」的反噬已经从「模型写简历写错学历」这种低级失误,蔓延到了「耗散关键行业的人工审核资源」层。\n\n## 几个值得记住的判断\n\n一是「AI slop」并不是一个孤立的趣味词——它是已经成为基础设施运营成本的噪声。Google 这类公司每年要消耗数千万美元在漏洞奖励支出上,而无效报告把预算和工程师注意力双重锁死,这是单纯加钱无法解决的。\n\n二是开源协作最核心的资产是「提交者信誉」,AI 让零信誉的提交瞬间伪装成有信誉,这个伪造不是技术问题,是社会协议问题。Google 用暂停作为短期止血、承诺 2027 年一季度给出系统性方案,这本身就说明现有 anti-spam 机制(lint、CWE 比对、人工审核)在 LLM 面前不够用了。\n\n三是这次冻结的反向推力会回灌到上游——AI 安全工具厂商接下来要回答的问题不是「能不能更快挖到洞」,而是「怎么在报告层面证明这个洞是我认真挖出来的」。这意味着 AI 安全赛道的产品形态会被迫从「发现」转向「可审计地报告」。\n\n对 AI 行业普通从业者来说,这件事离自己没那么远:你今天用的某个 LLM 工具也许不是漏洞源头,但是当一个 npm 包被无效 CVE 报告灌水后,真正修复该包的维护者反而没时间做他本来该做的事——这种延迟会以一种看不见的方式,沿依赖链传到下游。\n\n## 一个还没人回答的问题\n\n暂停产品漏洞提交之后,Google 会用什么机制替代原来的人工筛查?单纯加 CAPTCHA 把研究员挡回去是不可能的,这种机制会被真正有能力的白帽视为不友好;而完全依赖 LLM 评分又会回到同样的幻觉问题。Google 说 2027 一季度会更新,但愿那时候给出的不是另一种形式的「暂停就是把问题藏起来」。","https:\u002F\u002Fwww.helpnetsecurity.com\u002F2026\u002F10\u002F05\u002Fgoogle-ai-generated-vulnerability-reports-pause","2ea3b602-f810-4167-b1ea-b56c5de103f8",[11,15,18],{"id":12,"name":13,"slug":13,"description":14,"color":14},"c33b1bbc-d6ce-4f61-9d5d-1a0704a6a09b","ai-policy",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":19,"name":20,"slug":20,"description":14,"color":14},"02bfdc0f-5e23-4fbb-b31a-7cc03cdba1ee","security",[22],{"id":23,"lang":24,"title":25,"summary":26,"content":27},"207007de-fc74-4629-be4f-d6c3583e7651","en","Google pauses OSS VRP, citing AI bug slop","Google froze OSS VRP product submissions over a flood of invalid AI reports. Other open-source security programs see the same AI slop pressure.","On October 1, 2026, Google froze the product-vulnerability intake of its OSS VRP bug bounty program. The official justification is blunt: AI-generated reports flooded in, the overwhelming majority were invalid, and triaging them became the work itself. This is the latest, and most quantitatively clear, case of \"AI slop\" biting back at an industry that has been happy to ride the same tooling wave.\n\n## What OSS VRP paused and what it kept\n\nGoogle's pause is narrow: only \"product vulnerabilities\" are closed, submissions sent before October 1 are not affected, supply-chain attack reports continue to be accepted, and the Cloud VRP plus Patch Rewards Program stay open. In other words, this is the slot most exposed to low-quality AI submissions being closed, with a commitment to return with an updated program in Q1 2027. The official rules page now shows no reward table for product vulnerabilities across the four project tiers (OT0 through OT3), which is effectively a hard close of that entry point.\n\n## Why AI reports became unbearable\n\nOSS VRP covers projects like Go, Angular, and Protocol Buffers. Producing a real submission used to require real skill: actually locating the code defect, design bug, or supply-chain component issue. Large language models and automated bug-hunting scripts have compressed the cost-skill ratio to almost zero, anyone can ask an LLM to scan a codebase and emit a report that looks structurally legitimate. Help Net Security covered the warning signs in May; Google's pause is the formal answer to months of complaints from open-source maintainers already drowning in screening work.\n\nThe cost shows up as engineer time spent determining whether a given submission is a real vulnerability or a model hallucinating at an unfamiliar boundary. The end result is that the most serious bugs end up waiting in a queue while maintainers are stuck on triage, the core trust mechanism of open-source security collaboration gets quietly spent.\n\n## It is not just Google\n\nFour other players are visibly under the same pressure:\n\n- **arXiv** began enforcing a two-submissions-per-month cap on October 1, citing a wave of high-density, near-automated submissions.\n- **Internet Bug Bounty (IBB)** paused new submissions on March 27, 2026; the HackerOne-managed program that historically paid curl's bounties still shows as paused as of October 6 in the API.\n- **Linux kernel** maintainers publicly said they were \"completely overwhelmed,\" with the most recent release pushing the CVE count to a record 2,000 vulnerabilities, mostly produced by AI-assisted bug hunters.\n- **Intel** suspended its bug bounty program, which paid up to $100,000 per flaw. Intel has not officially confirmed AI-generated reports as the reason, but experts widely treat that as the likely explanation.\n\nThe implication is that \"AI is reshaping research and engineering workflows\" no longer stops at funny model failures. It is now measurably eating the human review budget that holds key open infrastructure together.\n\n## A few takeaways\n\nFirst, \"AI slop\" is not a joke label, it has crossed into being an operations cost on the books. Programs like these burn tens of millions of dollars a year across the industry on vulnerability rewards, and the wasted-report load burns that budget twice: once on rewards that never should have been paid out, and again in engineering attention that never lands on real bugs. Throwing more money at it does not solve it.\n\nSecond, the core asset of open-source collaboration is submitter reputation, and AI has made it trivial to spoof a credible-looking submission with zero reputation behind it. This is not a technical problem, it is a social-protocol problem. Using a pause as short-term triage while promising a Q1 2027 redesign tells you the existing anti-spam mechanisms (lint, CWE matching, manual review) are no longer adequate against LLM-generated volume.\n\nThird, the backlash pressure will run upstream. AI security vendors will increasingly be asked not just \"can you find bugs faster\" but \"can you credibly prove this report came from real work.\" This pushes the AI security product category from \"discovery\" to \"auditably attributed reporting.\"\n\nFor a normal engineer working on LLM products, this is not an abstract concern: every inflated CVE submission against an npm package translates into maintainer hours burned that should have gone toward real fixes. That delay travels down the dependency chain in ways nobody can easily track.\n\n## One open question\n\nOnce product-vulnerability submissions are paused, what does Google plan to replace the current human triage with? Adding CAPTCHAs is out, that approach would alienate the very skilled researchers the program needs. Relying entirely on LLM scoring reproduces the same hallucination problem. Google says Q1 2027. Hopefully the answer is not just \"pause as a way to hide the problem.\"","google-freezes-oss-vrp-ai-bug-slop","2026-10-11T02:00:00Z","2026-10-11T11:05:23.639725Z","2026-10-11T11:05:23.639735Z",true,"agent",135,[36,45],{"slug":37,"tag_slug":37,"title_zh":38,"title_en":39,"intro_zh":40,"intro_en":41,"id":42,"is_active":32,"created_at":43,"modified_at":44},"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":46,"tag_slug":46,"title_zh":47,"title_en":48,"intro_zh":49,"intro_en":50,"id":51,"is_active":32,"created_at":52,"modified_at":53},"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":55},[56,61,66,71,76,81],{"id":57,"title":58,"news_slug":59,"published_at":60},"488f30b2-34b1-48e4-b67f-9187f15c1a71","AI 抢先挖出 PHP 漏洞，披露窗口先被自己人捅破","openai-php-rce-coordinated-disclosure","2026-10-21T06:00:00+00:00",{"id":62,"title":63,"news_slug":64,"published_at":65},"c2c96e3a-cc8d-4970-8d9d-e9584bbc9237","WhatsApp 帮助页藏 102 块 LLM 暗指令","whatsapp-hidden-llm-instructions-help-pages","2026-10-19T16:00:00+00:00",{"id":67,"title":68,"news_slug":69,"published_at":70},"14a3cf67-03ef-4f85-a6a0-d809bfbfee04","Linux 内核不再要求 AI 辅助标注","linux-maintainers-drop-assisted-by-2026","2026-10-19T02:00:00+00:00",{"id":72,"title":73,"news_slug":74,"published_at":75},"de7af8cd-8c1c-4872-a0ff-ba2ca81ed5cc","Open Source Summit Europe 现场:Linux 阵营亲 AI,BSD 反 AI 阵营冲上 12 万美元捐款","oss-eu-prague-linux-vs-bsd-ai","2026-10-17T01:00:00+00:00",{"id":77,"title":78,"news_slug":79,"published_at":80},"b514c2a2-1d99-4666-8e78-588bb08e2979","Google 冻结 OSS VRP:AI 报告把维护者压垮","google-oss-vrp-ai-slop-pause","2026-10-12T02:00:00+00:00",{"id":82,"title":83,"news_slug":84,"published_at":85},"7809c36c-7ae0-409e-bb68-4c977b5f2a41","AI 幻觉报告泛滥 谷歌冻结开源漏洞悬赏计划","google-oss-vrp-pause-ai-slop-reports","2026-10-06T03:00:00+00:00"]