[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-chatgpt-eu-vlop-dsa-first-ai-platform-rules":3,"topics-all":38,"news-related-2fc4f696-a697-498c-b9d5-28250bfeaa79":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},"2fc4f696-a697-498c-b9d5-28250bfeaa79","ChatGPT 进欧盟 VLOP 名单:OpenAI 第一次要为生成式 AI 内容负全责","欧盟委员会本周正式将 ChatGPT、Reddit、Roblox 列为《数字服务法》项下的「超大型在线平台」(VLOP),三者在欧盟月活均已突破 4500 万。OpenAI 必须在 12 月底前履行 DSA 全部合规义务,违规将被处以最高全球营收 6% 的罚款,这是生成式 AI 首次完整落入欧盟平台监管框架。","## 背景:VLOP 是什么、为什么这次不一样\n\n欧盟《数字服务法》(DSA)把月活超过 4500 万的服务定义为「超大型在线平台」(Very Large Online Platforms, VLOP)。一旦戴上 VLOP 的帽子,平台要承担内容审核、未成年人保护、广告库公开、风险评估等一系列额外义务,违规最高可被处以全球年营收 6% 的罚款——对 OpenAI 这种年化数十亿美元体量的公司,理论罚单可以冲到数十亿美元的量级。\n\n过去 VLOP 名单上主要是社交媒体(比如 X、Meta 系)与电商(AliExpress)。这次新增的三个名字,只有 Reddit 和 Roblox 算是传统平台。ChatGPT 是名单上**第一个**生成式 AI 对话产品,这是 DSA 第一次以完整平台监管的口径对待一个 LLM 应用,而不是把它当作单纯的模型或工具来管。\n\n## 事件核心:欧盟的认定逻辑\n\n欧盟委员会周一公开了三家的认定决定,科技事务负责人 Henna Virkkunen 在声明中写道:\"ChatGPT、Reddit 和 Roblox 将在欧盟接受更高标准的审视和问责,这与它们对欧盟公民和社会的影响力相称。\"\n\n触发认定的是 4500 万欧盟月活的门槛。三家都在今年先后突破这一阈值。OpenAI 发言人确认公司正在准备满足额外的合规要求;Roblox 自称是\"欧盟第一家达到 VLOP 门槛的游戏平台\";Reddit 在截稿前未回应媒体询问。\n\n合规截止日是**今年 12 月底**。也就是说,OpenAI 只有不到四个月的时间把内部的内容审核、广告投放、风险评估体系改造到符合 DSA 的程度。\n\n## 为什么这一刀对 OpenAI 格外重\n\nDSA 原本针对的是社交平台的信息流和评论区。对于以「用户提问、模型回答」为产品形态的 ChatGPT,这些义务要被重新映射一遍。例如:\n\n- 内容审核义务:模型输出是否算「平台发布的非法内容」?如果用户用 ChatGPT 生成儿童性侵材料,OpenAI 是否承担和 X 转发恐怖主义视频同等的责任?\n- 广告库公开:ChatGPT 已经在免费版里插入广告,这些广告的定向逻辑、用户画像如何透明化?\n- 风险评估:需要提交年度系统性风险评估报告,涵盖非法内容、未成年人保护、对公民话语的影响——这套体系原本是为社交网络设计的。\n\n去年欧盟已经对 X 的 AI 聊天机器人 Grok 启动过 DSA 调查。但 Grok 在 X 体系内是子产品,这次对 ChatGPT 的认定是直接以平台为单位,范围更完整。\n\n## 与 AI Act 的关系:DSA 是即时刀、AI Act 是长期刀\n\n欧盟手上现在有两把针对 AI 的刀:AI Act 已经在分阶段落地,管的是模型的开发与部署;而 DSA 这一次伸向了**用户接触产品的那个界面**。OpenAI 不能只满足 AI Act 的透明度义务,还要满足 DSA 的平台级义务。\n\n更复杂的还在后面。欧盟正在依据竞争规则调查 Google 在 AI Overviews 中的使用方式,也在同步推进 AI Act 的执法。两套法律叠加,意味着任何一家在欧盟运营的 AI 公司都要同时应付模型层和应用层的监管。\n\n## 我的判断:这次不只是罚款威胁,而是合规形态重写\n\n短期看,OpenAI 不会因为这次认定就做出激进的产品改动——主要任务是合规文档和流程改造。但中长期,这件事会重写生成式 AI 产品的几个默认假设:\n\n1. **模型输出不再是「用户行为」,而是「平台内容」**。这意味着 OpenAI 需要为每条模型输出建立可追溯的责任链,而不仅仅是优化拒绝率和 RLHF。\n2. **广告变现路径被收紧**。DSA 要求 VLOP 公开广告库并禁止基于敏感属性的定向,这对 ChatGPT 试图把广告铺进回答里的策略是一记重击。\n3. **欧盟外的溢出效应**。和此前 AI Act 第 50 条文本水印的情况类似,一旦 OpenAI 为欧盟单独改造了产品,这部分合规能力最终会下沉为「全球默认」。\n\n说到底,这次认定不是让 OpenAI 担心下一张罚单,而是让生成式 AI 第一次被当作「平台」来治理。模型的归模型,平台的归平台——这条线,欧盟这次划清楚了。","https:\u002F\u002Farstechnica.com\u002Ftech-policy\u002F2026\u002F08\u002Fchatgtp-and-reddit-now-face-eus-toughest-online-safety-rules\u002F","2af9d198-9418-4f26-85e4-4a8f3eede35a",[11,15,18,21],{"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},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"42e59a88-7795-47dc-a334-ef1e72c24347","openai",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"eb66dbc0-5aa7-4558-8dff-e2aa07f0aed2","en","ChatGPT Joins EU VLOP List: First AI Under Full Platform Rules","The EU designated ChatGPT, Reddit, and Roblox as Very Large Online Platforms under the DSA. OpenAI must comply by year-end or face fines up to 6% of revenue.","## Background: What VLOP Means and Why This Time Is Different\n\nThe EU's Digital Services Act (DSA) defines any service with more than 45 million monthly users in the EU as a Very Large Online Platform (VLOP). Once a service gets the VLOP label, it must comply with a long list of additional obligations — content moderation, minor protection, ad library disclosure, systemic risk assessment, and more. Non-compliance can trigger fines of up to 6% of global annual revenue, which for a company like OpenAI, with revenue in the multi-billion range, would translate into potential penalties of billions of dollars.\n\nThe VLOP list until now has been populated mostly by social media platforms (such as X and the Meta family) and marketplaces (AliExpress). Of the three new entries this round, only Reddit and Roblox fit the conventional platform mold. ChatGPT is the **first** generative AI conversational product ever placed on the VLOP list. This is the first time the DSA has treated an LLM application under the full platform-regulator lens, rather than as a model or tool.\n\n## Core of the Decision: How Brussels Got There\n\nThe European Commission made the designation public on Monday, with tech chief Henna Virkkunen stating: \"ChatGPT, Reddit, and Roblox will now be held to a higher standard of scrutiny and accountability in the European Union, in line with their large impact on our citizens and society.\"\n\nThe trigger was the 45 million EU monthly user threshold, which all three services have now crossed. An OpenAI spokesperson confirmed the company is preparing to meet the additional compliance requirements. Roblox called itself \"the first gaming platform to reach this milestone in the EU.\" Reddit did not respond to media inquiries by the time of publication.\n\nThe compliance deadline is **the end of December this year**. In other words, OpenAI has under four months to reshape its internal content moderation, advertising, and risk assessment systems to DSA standards.\n\n## Why This Cut Cuts OpenAI Especially Deep\n\nThe DSA was originally designed around social media feeds and comment sections. For a product whose primary interface is \"user prompts, model answers,\" each obligation has to be remapped. For example:\n\n- **Content moderation obligation**: Is a model's output \"illegal content published by the platform\"? If a user uses ChatGPT to generate child sexual abuse material, does OpenAI bear the same responsibility as X hosting terrorist videos?\n- **Ad library disclosure**: ChatGPT has already started inserting ads in the free tier. How must the targeting logic and user profiles be made transparent?\n- **Risk assessment**: OpenAI must now file annual systemic risk assessment reports covering illegal content, minor protection, and impact on civic discourse — a system originally built for social networks.\n\nBrussels has already opened a DSA investigation into X's AI chatbot Grok, but Grok sits inside the X platform as a sub-product. This designation treats ChatGPT directly as a platform, in a way that is broader and more complete.\n\n## Relationship with the AI Act: DSA Is the Immediate Blade, AI Act the Long Arc\n\nBrussels is currently wielding two knives against AI: the AI Act, which has been phasing in and governs model development and deployment; and the DSA, which this time reaches into the **user-facing interface**. OpenAI cannot satisfy the AI Act's transparency obligations alone — it must also meet DSA's platform-level requirements.\n\nMore complexity is coming. The EU is investigating Google's use of AI Overviews under competition rules, and is enforcing the AI Act in parallel. Two legal regimes stacked together mean any AI company operating in the EU now has to satisfy both the model layer and the application layer of regulation.\n\n## My Take: This Is Not Just a Fine Threat — It Is a Rewrite of Compliance Shape\n\nIn the short term, OpenAI won't make radical product changes because of this designation — the main task is documentation and process transformation. In the medium to long term, however, this event will rewrite several default assumptions of generative AI products:\n\n1. **Model output is no longer \"user behavior\" but \"platform content.\"** This means OpenAI needs a traceable accountability chain for every model output, not just refusal rate and RLHF optimization.\n2. **The ad monetization path is being tightened.** The DSA requires VLOPs to disclose ad libraries and prohibit targeting based on sensitive attributes — a heavy blow to ChatGPT's strategy of placing ads inside answers.\n3. **Spillover effects outside the EU.** Similar to the earlier case of AI Act Article 50 text watermarks, once OpenAI rebuilds its product for the EU alone, that compliance capability will eventually sink into a \"global default.\"\n\nIn short, this designation does not have OpenAI worried about the next fine — it is the first time generative AI has been governed as a \"platform.\" Models belong to models, platforms belong to platforms — that line, the EU has now drawn clearly.","chatgpt-eu-vlop-dsa-first-ai-platform-rules","2026-09-05T07:00:00Z","2026-09-05T05:04:21.800839Z","2026-09-05T05:04:21.800849Z",true,"agent",134,[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},"1e43b4fc-39fe-4cac-aaf9-57f82d5c0311","AI 公司与数学界「错位」:两个月三次刷屏,把同行评审甩在身后","ai-math-severe-misalignment-fields-medal","2026-09-15T10:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"982c5e1e-5274-442e-9237-abaf39e8ee3c","25 位菲尔茨奖得主联名公开信:AI 解题竞赛正在伤害数学","fields-medalists-ai-misalignment-math","2026-09-13T13:07:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"6b349d2c-3d03-4cb9-8f47-63e8c288d0db","美方三机构联合指控六家中国 AI 企业系统性蒸馏美国模型","us-accuses-six-chinese-ai-firms-of-distillation","2026-09-10T01:08:40+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"d17a841b-abca-46e0-80e4-d955f1c837ba","亚马逊 VGT3 仓库曝光:一天拆掉上千本书,只为给 AI 模型喂语料","amazon-vgt3-warehouse-ai-training-books","2026-09-07T03:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"1942b07b-f794-42b1-b944-ca6b32d4ae16","四大 AI 模型同日集体掉线:OpenAI\u002FClaude 官方确认,Gemini\u002FGrok 表面沉默","four-ai-models-overlapping-outage-sept-2026","2026-09-06T08:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"390c2437-4e4f-45ec-8270-67c5bfa4fa47","ChatGPT、Claude、Grok、Gemini 罕见同时下线,周四早晨全球 AI 集体失声","chatgpt-claude-grok-gemini-thursday-outage","2026-09-05T06:00:00+00:00"]