[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-google-oss-vrp-ai-slop-pause":3,"topics-all":35,"news-related-b514c2a2-1d99-4666-8e78-588bb08e2979":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},"b514c2a2-1d99-4666-8e78-588bb08e2979","Google 冻结 OSS VRP:AI 报告把维护者压垮","2026 年 10 月 1 日起,Google 暂停接收 OSS VRP 的产品漏洞提报,称「自动化提报数量显著上升,其中绝大多数无效」,新规要到 2027 年一季度才更新;这与 curl、Intel 等项目今年早些时候因 AI 提报洪水关闭漏洞奖励的趋势一致。","2026 年 10 月 1 日,Google 悄悄把自己最大的开源漏洞奖励计划 OSS VRP 暂停了一年——理由不是钱,也不是合规,而是被 AI 生成的「看似合理的」漏洞报告淹没了。Google 漏洞奖励小组在 X 平台发文写得很直接:「之所以暂停,是因为自动化提报的数量显著上升,其中绝大多数是无效的」。对于一个成立于 2022 年、覆盖 Go、Angular、Protocol Buffers 等自家开源核心项目的奖励计划来说,这等于把信息通道暂时关掉。\\n\\n## 暂停的不是发现,是入口\\n\\nOSS VRP 的产品漏洞通道从 10 月 1 日起停止接收新报告,提交方式并没有消失,只是被引导到 Cloud VRP 的对应分支,以及 Google 自家的 Patch Rewards Program——但这两条路主要面向「主动修复」而不是「提交漏洞」。Help Net Security 报道援引 Google 的官方规则页,OS VRP 的四条项目等级(从旗舰 OT0 到低优先级 OT3)的奖励表里,「产品漏洞」一栏目前没有金额。换句话说,Google 仍在为漏洞修复和现有研究付钱,但不再为「报告一个我从未存在的产品漏洞」付钱。\\n\\n云端的 Cloud VRP 仍对部分 Google Cloud 仓库开放,10 月 1 日之前的报告也照常处理;Google 承诺 2027 年一季度给出更新方案。这意味着这不是「撤项」,而是一次「重设入口门」的版本切换。\\n\\n## 不是孤立事件:AI 把报告的发件箱和收件箱都打通了\\n\\nOSS VRP 并不是第一个受害者。Malwarebytes 的报道把上下文摆开:curl 在 2026 年初关闭了 HackerOne 上的漏洞奖励,原因是小团队扛不住大量低质量且大量由 AI 生成的提报;Intel 在 Intigriti 上的新漏洞奖励计划同样悄悄停掉了金钱悬赏,据 Phoronix 报道也是因为 AI 报告的洪流。当 OpenAI、Claude 这类代码模型把「写一份格式完整的 CVE 报告」的成本压到接近零,提交端的生产力天花板被拿掉,而维护者端的审核、复现、去重成本并没有同等下降。\\n\\n这种「发送成本暴跌、接收成本不变」的不对称,正是经典的 spam economics:写一份看上去合理的报告便宜到可以投一百份,但维护者接到每一份都还要走一遍构建、复现、定位代码的完整流程。于是出现了「无效报告淹没有效报告」的结构性问题,Google 选的是直接关停一年,curl 选的是退出,Intel 选的是只留非金钱奖励。\\n\\n## 这次不一样:门槛低到让 AI 写出的漏洞看起来很真\\n\\n早期代码 LLM 写出的报告更容易被人一眼识破——错位的 API 名、不存在的调用栈、前言不搭后语的复现步骤。现在不同了。Cursor、Claude Code 这类 Agent 工具能直接拉代码仓库、跑构建命令、读 traceback,生成的结构化报告在专业评审眼里也要花几小时才能定位出无效。Malwarebytes 的分析师 Pieter Arntz 把它说成「披露经济学早已存在的漏洞,只不过 AI 把它放大了」——他给出的对策很直接:让 AI 接管提报的第一轮过滤,只有被 AI 验证为有效的报告才转给工程团队。这听起来像把守门人也换成机器人,但比「关停一年」的实际效果更好。\\n\\n这意味着当 OSS VRP 在 2027 年一季度重新开放时,它大概率不会变回 2026 年之前的版本。最可能的剧本是:要求强制 PoC、要求最低复现成功率、拉高最低项目等级门槛、引入 dedup 工具链,以及 Google 内部用一套 AI 流水线先卡一层再放给人类评审。这条路径会很别扭,但在「拒绝低质量报告」和「不打压真正研究者」之间,几乎找不到不别扭的解。\\n\\n## 真正的代价会落在中型开源项目\\n\\nGoogle OSS VRP 还算撑得起——它有专门团队,有 Cloud VRP 兜底,有 2027 年再开张的时间窗口。但这件事的连带效应会落在别处:那些没有专门安全团队、靠社区 HackerOne 维护的中型开源项目,面对「一年后再开」的 OSS VRP 没有第二选择。它们要么花钱请商业审核方,要么自己雇一个 AI 评审流水线,要么暂时把漏洞披露门槛抬到不合理的程度。换句话说,Google 这一年的暂停,更像一次把漏洞披露经济的「成本曲线」一次性摊在桌面上的动作——LLM 让发件箱几乎免费,买这种便宜的人不只 Google 的维护者,任何还在公开运营漏洞披露的项目都会感受到冲击。\\n\\n所以最值得记住的不是 Google 关停了哪一项,而是:当 AI 让无意义的提报「成本为零、维护者成本不变」时,安全生态是被这一项成本结构逼出来的。OSS VRP 暂停,只是把这种结构问题摆到了最受关注的位置。\\n\\n参考资料:Help Net Security《AI slop submissions force Google to freeze its open-source bug bounty》(https:\u002F\u002Fwww.helpnetsecurity.com\u002F2026\u002F10\u002F05\u002Fgoogle-ai-generated-vulnerability-reports-pause\u002F)、Malwarebytes《Google pauses open source bug bounty program after rise in AI submissions》(https:\u002F\u002Fwww.malwarebytes.com\u002Fblog\u002Fnews\u002F2026\u002F10\u002Fgoogle-pauses-open-source-bug-bounty-program-after-rise-in-ai-submissions)、Solidot 转载《因涌入大量 AI 报告 Google 冻结其 Bug 悬赏计划》(https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85538)。","https:\u002F\u002Fwww.helpnetsecurity.com\u002F2026\u002F10\u002F05\u002Fgoogle-ai-generated-vulnerability-reports-pause\u002F","8ab50c9c-8ead-42bc-82ae-62614ea27632",[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},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"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},"6bc836c1-645c-4450-922c-969b33c7f0fe","en","Google freezes OSS VRP: AI reports overwhelm maintainers","Google stopped accepting OSS VRP product-vulnerability submissions on October 1, citing a flood of invalid AI-generated reports; Q1 2027 redesign.","On October 1, 2026, Google quietly paused one of its largest open-source bug-bounty programs, OSS VRP, for nearly a year — not over budget or compliance, but because AI-generated \"plausible-looking\" vulnerability reports drowned it. The official Google VRP account put it bluntly on X: \"This pause is due to a significant rise in automated submissions, the vast majority of which are not valid.\" For a program launched in 2022 that paid researchers for flaws in Go, Angular, Protocol Buffers, and other core open-source Google projects, the move is effectively closing the inbound channel — at least temporarily.\\n\\n## The entry is paused, not the practice\\n\\nOSS VRP stopped accepting new product-vulnerability reports on October 1. Submission is not gone, but it has been redirected: most new reports are now steered into the Cloud VRP for select Google Cloud repositories, or into Google's Patch Rewards Program — both of which compensate fixes more than raw vulnerability submissions. Help Net Security reported that Google has effectively stripped payouts for the \"product vulnerabilities\" column of the OSS VRP reward table across all four project tiers (OT0 flagship down to OT3 low-priority). In short, Google will keep paying for fixes, but no longer for \"report a product vulnerability that may not exist.\" Cloud VRP remains open, pre-October-1 reports continue, and Google has committed to a Q1 2027 redesign. The move isn't a withdrawal; it's an entry-gate rewrite.\\n\\n## Not an isolated event: AI widens both the inbox and the outbox\\n\\nOSS VRP is not the first casualty. Malwarebytes laid out the wider landscape: curl terminated its HackerOne bounty program in early 2026 because a tiny security team could no longer absorb the flood of mostly AI-generated, low-quality reports. Intel similarly ended cash payouts on its Intigriti bug-bounty program, reportedly for the same reason, per Phoronix. When code-capable models from OpenAI, Anthropic, and others cut the cost of producing a well-formatted CVE-style report to near-zero, the supply-side productivity ceiling got removed — but the receiver-side cost of building, reproducing, and deduping every submission did not.\\n\\nThat asymmetric drop — sender cost collapses, receiver cost unchanged — is classic spam economics. Concretely, writers can flood 100 submissions for the price of one, while reviewers still pay the full cost per submission. The result is invalid reports drowning out the valid ones. Google chose to shut the door for a year; curl chose to leave; Intel chose non-monetary recognition only.\\n\\n## The bar is set so low that AI-crafted vulns look real\\n\\nReports from early code LLMs were easy to triage — wrong API names, fabricated stack traces, garbled reproduction steps. That edge has eroded. Agent tools such as Cursor and Claude Code can clone a repository, run its build, read the traceback, and produce a structured vulnerability narrative that even seasoned reviewers need hours to disprove. Malwarebytes analyst Pieter Arntz framed this as an old weakness in disclosure economics that AI has now exposed: the cost of filing a plausible report has fallen sharply, while the cost of proving or disproving it remains human-intensive. His suggested fix is direct — let AI handle intake triage and only forward validated reports to engineers. Putting a robot at the gate sounds awkward, but it is operationally better than shutting the program for a year.\\n\\nIn practice, when OSS VRP reopens in Q1 2027 it almost certainly will not be the 2025 version again. The most likely redesign lands somewhere between: mandatory proof-of-concept evidence, raised minimum tier thresholds, deduplication tooling, and an internal AI filter layer ahead of human review. None of those are clean answers, but the alternative — letting the queue keep piling up — is worse.\\n\\n## The real cost lands on the open-source middle\\n\\nGoogle OSS VRP has dedicated headcount, Cloud VRP fallback, and a one-year runway to redesign. Most projects do not. Mid-sized open-source projects without dedicated security staff, which rely on HackerOne or community disclosure, now lose one of their natural escalation paths. They will either pay a commercial triage vendor, stand up their own AI review line, or quietly raise their disclosure threshold. In other words, Google's pause is also a stress-test of disclosure economics when the marginal cost of filing falls to near-zero. The total maintenance burden shifts onto smaller shoulders.\\n\\nThe lesson worth keeping is not which program Google paused. It is this: when LLM tooling drives submission costs toward zero while reviewer costs remain human-scale, the cost structure itself forces a redesign of public vulnerability disclosure — and Google's one-year pause is just the highest-visibility signal of that broader shift.","google-oss-vrp-ai-slop-pause","2026-10-12T02:00:00Z","2026-10-12T05:14:22.669537Z","2026-10-12T05:14:22.669552Z",true,"agent",73,[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},"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":62,"title":63,"news_slug":64,"published_at":65},"ac04da32-378f-403c-83c9-4cf2e29948e3","Claude半小时破解NSA曲线种子25年悬案","claude-nist-curve-seeds-lrand48","2026-10-17T19:20:00+00:00",{"id":67,"title":68,"news_slug":69,"published_at":70},"db858586-4427-4480-bfe0-fa7747ab6d10","Google 冻结 OSS VRP：AI 报告淹没开源赏金","google-freezes-oss-vrp-ai-bug-slop","2026-10-11T02:00:00+00:00",{"id":72,"title":73,"news_slug":74,"published_at":75},"f093935e-e23c-4766-88d6-9302a370be2a","Gemini 套餐洗牌:免费版限 Flash-Lite","gemini-deep-think-ai-pro-tier-restructure","2026-10-09T00:00:00+00:00",{"id":77,"title":78,"news_slug":79,"published_at":80},"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",{"id":82,"title":83,"news_slug":84,"published_at":85},"7d3de6b6-1faf-47fc-a513-62ee8e731aad","CNNIC 2026 报告:7 亿用户、2185 EFLOPS 算力、调用前六全是中国","cnnic-2026-china-generative-ai-700m-users-2185-eflops","2026-09-29T06:00:00+00:00"]