[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-detail-361":3,"$fOHoMwCnpy0REzFpuPjR7qV926ifzaOXjvedDr3bPDiQ":11,"news-list":135},{"id":4,"title":5,"excerpt":6,"content":7,"image":8,"date":9,"category":10},"361","AI”清朗”专项行动——当”数字泔水”成为政策术语：中国AI监管进入”定义时代”","“清朗·整治AI应用乱象”专项行动共分14类问题，覆盖从模型训练、语料安全、数据投毒、内容标识，到AI生成内容、AI水军、AI产品违规的全链条。","\u003Csection>\u003C/section>\n\u003Ch2>\u003Cstrong>一、一次全国性的AI监管框架升级\u003C/strong>\u003C/h2>\n\u003Cp>2026年4月30日，中央网信办印发通知，在全国范围内部署开展为期4个月的“清朗·整治AI应用乱象”专项行动。\u003C/p>\n\u003Csection>\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"wp-image-368 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143102482-300x90.png\" alt=\"\" width=\"464\" height=\"139\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143102482-300x90.png 300w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143102482-1024x306.png 1024w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143102482-768x230.png 768w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143102482-1536x459.png 1536w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143102482-2048x612.png 2048w\" sizes=\"auto, (max-width: 464px) 100vw, 464px\" />\u003C/section>\n\u003Cp style=\"text-align: right;\" data-pm-slice=\"0 0 []\">图源 ｜ 中央网信办\u003C/p>\n\u003Cp>这与以往常规的“打击网上不良信息”不同——过往十余年的“清朗”系列行动，整治对象大多是既有的网络乱象（虚假信息、低俗内容、水军等），而这一次，监管的对象是AI技术本身，以及AI技术催生出的全新内容生产形态。\u003C/p>\n\u003Cp>从整治框架来看，这次行动分为两个阶段，共14类问题，覆盖从模型训练、语料安全、数据投毒、内容标识，到AI生成内容、AI水军、AI产品违规的全链条。\u003C/p>\n\u003Cp>这是中国AI监管史上首次出现如此系统化的全链条\u003Cstrong>整治清单\u003C/strong>。\u003C/p>\n\u003Cp>上海网信办5月中旬的执行方案之所以值得关注，并不是因为上海“率先”做了什么，而是因为上海是中国AI产业最集中的城市之一——商汤、MiniMax、阶跃星辰等头部AI公司的总部均在上海，小红书的AI内容生态、B站的AI视频创作、携程的AI客服、拼多多的AI电商应用，也都是上海属地平台。上海怎么执行，很大程度上会影响全国AI企业的合规基准线。\u003C/p>\n\u003Ch2 data-pm-slice=\"0 0 []\">二、“数字泔水”：一个被官方定义的AI内容新概念\u003C/h2>\n\u003Cp data-pm-slice=\"0 0 []\">“数字泔水”这个词出现在中央网信办的原始通告中，用来定义第二阶段的整治对象之一：“批量生成发布逻辑混乱、价值空洞，或带有错误价值观、扭曲文化认知等不良倾向”的AI内容。\u003C/p>\n\u003Cp>\u003Cstrong>这个词\u003C/strong>\u003Cstrong>的重点\u003C/strong>\u003Cstrong>在于，它不是在定义“违法”，而是在定义“无价值”。\u003C/strong>\u003C/p>\n\u003Cp>传统的网络治理框架里，“不良信息”有明确的法律边界：虚假信息、淫秽色情、暴力恐怖、煽动仇恨等，这些在《网络安全法》《网络信息内容生态治理规定》里都有对应条款。\u003C/p>\n\u003Cp>但“数字泔水”指向的是一种未必违法、但质量\u003Cstrong>很低甚至无\u003C/strong>\u003Cstrong>价值的内容\u003C/strong>——它可能是语法通顺、不触发任何敏感词，甚至看起来“正常”的AI生成内容。但它没有人类的心智参与，没有信息增量，没有文化价值，纯粹是为了占据用户的注意力而存在的网络垃圾。\u003C/p>\n\u003Cp>\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-366 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143101270-300x179.jpeg\" alt=\"\" width=\"300\" height=\"179\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143101270-300x179.jpeg 300w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143101270-1024x611.jpeg 1024w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143101270-768x458.jpeg 768w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143101270.jpeg 1080w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" />\u003C/p>\n\u003Cp style=\"text-align: right;\">海外社交媒体上流行的AI”Shirmp Jesus”\u003C/p>\n\u003Cp style=\"text-align: right;\">图源 ｜ Facebook\u003C/p>\n\u003Cp>中央网信办选择用一个如此具象、甚至带有一点情绪色彩的词汇来给这类内容定性，传递了一个明确的监管信号：A\u003Cstrong>I内容治理的边界正从法律\u003C/strong>\u003Cstrong>层面\u003C/strong>\u003Cstrong>扩展到价值\u003C/strong>\u003Cstrong>层面\u003C/strong>\u003Cstrong>。\u003C/strong>不是只有“违法”的AI内容才需要管，“低质”的AI内容同样在被治理的范围内。\u003C/p>\n\u003Cp>这是一个重大的监管理念转变。如果这一思路被写入后续的常态化监管框架，意味着未来AI内容平台不仅要过滤违法违规信息，还要建立对AI生成内容质量的评估和治理能力——这比单纯的内容审核要复杂得多。\u003C/p>\n\u003Cp>\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-362 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059854-300x178.png\" alt=\"\" width=\"300\" height=\"178\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059854-300x178.png 300w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059854.png 720w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" />\u003C/p>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"3 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\n\u003Cp style=\"text-align: right;\">社交媒体流行的AI橘猫\u003C/p>\n\u003Cp style=\"text-align: right;\">图源 ｜ 知乎\u003C/p>\n\u003C/section>\n\u003C/section>\n\u003Ch2 data-pm-slice=\"0 0 []\">三、为什么是14类问题？拆解监管框架的应对逻辑\u003C/h2>\n\u003Ch3 data-pm-slice=\"0 0 []\">第一阶段：在“源头”筑坝\u003C/h3>\n\u003Csection>\n\u003Cp data-pm-slice=\"0 0 []\">第一阶段名为“AI应用服务典型违规问题”专项治理。仔细看治理清单，它实际上是在AI产业链的\u003Cstrong>上游\u003C/strong>设置监管闸门：\u003C/p>\n\u003C/section>\n\u003Csection>\n\u003Cul class=\"list-paddingleft-1\">\n\u003Cli>\u003Cstrong>大模型备案\u003C/strong>→ 管的是“谁在做AI”\u003C/li>\n\u003C/ul>\n\u003Cul class=\"list-paddingleft-1\">\n\u003Cli>\u003Cstrong>安全审核能力\u003C/strong>→ 管的是“做出来的AI安不安全”\u003C/li>\n\u003C/ul>\n\u003Cul class=\"list-paddingleft-1\">\n\u003Cli>\u003Cstrong>训练语料安全\u003C/strong>→ 管的是“AI学什么”\u003C/li>\n\u003C/ul>\n\u003Cul class=\"list-paddingleft-1\">\n\u003Cli>\u003Cstrong>数据投毒\u003C/strong>→ 管的是“有没有人故意让AI学坏”\u003C/li>\n\u003C/ul>\n\u003Cul class=\"list-paddingleft-1\">\n\u003Cli>\u003Cstrong>内容标识\u003C/strong>→ 管的是“AI生成的内容能不能被识别”\u003C/li>\n\u003C/ul>\n\u003Cul class=\"list-paddingleft-1\">\n\u003Cli>\u003Cstrong>技术滥用\u003C/strong>→ 管的是“AI被用来干什么坏事”\u003C/li>\n\u003Cli>\u003Cstrong>开源模型管理\u003C/strong>→ 管的是“开源社区的责任边界”\u003C/li>\n\u003C/ul>\n\u003Cp data-pm-slice=\"0 0 []\">这7类问题的共同点是：它们都是在内容产生之前设置的关卡。监管的逻辑非常清晰——如果上游的问题不解决，下游的内容乱象则永远清不完。\u003C/p>\n\u003Cp>其中有两个值得注意的条目：\u003C/p>\n\u003Cp>\u003Cstrong>AI数据投毒。\u003C/strong>官方定义是“通过篡改训练语料、伪造权威数据、使用\u003Cspan class=\"\">GEO技术\u003C/span>恶意营销等方式实施AI数据投毒”。\u003C/p>\n\u003Cp>GEO是指通过优化内容来操控AI搜索引擎的输出结果，其逻辑类似于传统SEO操控搜索引擎排名，但作用对象是AI而非人类。【大模型投毒】这个概念在学术界和产业界都还是前沿话题，能被写入中央层面的监管文件，可见当下收获的重视。\u003C/p>\n\u003Cp>\u003Ca class=\"normal_text_link mp_article_text_link\" href=\"https://mp.weixin.qq.com/s?__biz=Mzk4ODIxMzg1Mw==&mid=2247484793&idx=1&sn=6f42de4d8b5e99439756400bb64c30ae&scene=21#wechat_redirect\" target=\"_blank\" rel=\"noopener\" data-itemshowtype=\"0\" data-linktype=\"2\">揭秘GEO灰产：有人正批量给你的AI小助手塞小广告、数据投毒……\u003C/a>\u003C/p>\n\u003Cp>\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-363 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059205-300x196.jpeg\" alt=\"\" width=\"300\" height=\"196\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059205-300x196.jpeg 300w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059205-1024x669.jpeg 1024w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059205-768x502.jpeg 768w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143059205.jpeg 1080w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" />\u003C/p>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"3 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\n\u003Cp style=\"text-align: right;\">央视“3·15”晚会曝光AI投毒\u003C/p>\n\u003Cp style=\"text-align: right;\">图源 ｜ 央视\u003C/p>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Cp>\u003Cstrong>开源模型安全管理。\u003C/strong>魔搭社区（阿里巴巴旗下的开源模型平台）被纳入17家重点平台，其专项治理公告明确提到“对社区内托管的模型、数据集等核心资源展开全面安全审查”。\u003C/p>\n\u003Cp>这是一个标志性事件——中国的开源AI生态正在从“自由共享”进入“合规共享”阶段。这次专项行动将“开源模型安全管理”列为7类源头问题之一，意味着未来开源社区可能需要建立类似应用商店的审核机制：身份认证、内容审查、风险评估、应急处置。这将对未来的AI开源生态产生深远影响。\u003C/p>\n\u003Cp>\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-364 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100926-300x126.jpeg\" alt=\"\" width=\"300\" height=\"126\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100926-300x126.jpeg 300w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100926-1024x429.jpeg 1024w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100926-768x321.jpeg 768w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100926.jpeg 1080w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" />\u003C/p>\n\u003Csection>\n\u003Ch3 data-pm-slice=\"0 0 []\">\u003Cspan data-pm-slice=\"1 1 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"heading",{"tagName":"h3","attributes":{"class":"heading-3 ace-line old-record-id-XaMudm20EoEPlTx8GJNc1eoanoy","data-pm-slice":"0 0 []","style":"white-space: normal; margin: 0px; padding: 0px; box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml","level":3}]\">第二阶段：\u003C/span>\u003Cspan data-pm-slice=\"1 1 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"heading",{"tagName":"h3","attributes":{"class":"heading-3 ace-line old-record-id-XaMudm20EoEPlTx8GJNc1eoanoy","data-pm-slice":"0 0 []","style":"white-space: normal; margin: 0px; padding: 0px; box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml","level":3}]\">在”下游”清理\u003C/span>\u003C/h3>\n\u003C/section>\n\u003Cp data-pm-slice=\"0 0 []\">如果说第一阶段是在管“工厂”，第二阶段就是在管“产品”——也就是普通用户能直接看到的内容乱象。\u003C/p>\n\u003Cp>其中最具前瞻性的条目是“利用AI‘托管’从事网络水军活动”。官方定义是：“通过AI‘托管’技术运营账号，模拟真人批量注册、运营社交账号，批量生成发布低质同质文案。非法交易买卖账号。使用AI群控软件、社交机器人刷量控评，生成虚假流量数据，制造虚假舆论热点。”\u003C/p>\n\u003Cp>“\u003Cspan class=\"\">AI托管\u003C/span>”这个表述意味着，监管部门已经将AI驱动的自动化账号运营纳入打击范围——这不仅仅是内容问题，更是平台账号生态和流量真实性的问题。\u003C/p>\n\u003Cp>对社交媒体平台（微博、小红书、抖音、B站）来说，这意味着未来需要升级反bot能力，识别和打击由AI驱动的批量注册、内容发布、互动刷量行为。\u003C/p>\n\u003Csection>\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-365 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100272-203x300.jpeg\" alt=\"\" width=\"203\" height=\"300\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100272-203x300.jpeg 203w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100272-693x1024.jpeg 693w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100272-768x1135.jpeg 768w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100272-1039x1536.jpeg 1039w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143100272.jpeg 1080w\" sizes=\"auto, (max-width: 203px) 100vw, 203px\" />\u003C/section>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"3 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\n\u003Cp style=\"text-align: right;\">博主发布游戏星露谷物语相关视频，遭AI水军号评论\u003C/p>\n\u003Cp style=\"text-align: right;\">图源 ｜ 抖音\u003C/p>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection data-pm-slice=\"3 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\n\u003Csection>\n\u003Csection>\n\u003Csection>\n\u003Csection>\n\u003Csection>\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection>\n\u003Csection>\n\u003Csection>\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection>\n\u003Csection>\n\u003Csection>\n\u003Csection>\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Ch2 data-pm-slice=\"0 0 []\">四、合规声明——17家平台交出的不同“作业”\u003C/h2>\n\u003C/section>\n\u003Csection>同一份中央通知，17家上海属地平台交出了不同的“作业”。这背后折射出的是不同企业合规能力的结构性差异。\u003C/section>\n\u003Csection>\n\u003Ch3>AI模型公司：MiniMax、商汤、千问\u003C/h3>\n\u003C/section>\n\u003Csection>这类公司的核心产品是AI模型和AI应用，合规的重点在第一阶段——备案、语料、安全围栏、内容标识。MiniMax和商汤的公告高度标准化，套用了”严格对照两大阶段、十四类整治重点……全链条开展深度自查整改”的统一话术。千问的公告更详细一些，完整列出了两阶段共14类问题的定义。\u003Cstrong>为什么它们的公告看起来“模板化”？\u003C/strong>不是因为不重视，而是因为AI公司的合规动作相对“内聚”——主要是内部的产品和技术调整，不需要像内容平台那样对外公布具体的治理动作。它们的合规重点是模型备案、训练语料合规、输出内容标识，这些动作发生在技术层面，对外界而言就是”我们在整改”，很难像B站那样具体描述”把AI标识放到一级页面”。\u003C/section>\n\u003Csection>\n\u003Ch3>内容平台：B站、小红书、虎扑、Soul\u003C/h3>\n\u003C/section>\n\u003Csection>这类平台的合规重点是第二阶段——平台上由用户生成的AI内容怎么管。B站的公告最有可操作性：AI标识“提级至一级页面”、打击AI虚假配音/换脸误导/虚假新闻。虎扑和Soul则展现了不同的风格——虎扑“原文照搬”官方定义，Soul在技术层面最为深入，特别提到了GEO恶意营销和智能体窃取数据。\u003Cstrong>Soul提到的“智能体技术窃取用户数据或账户密钥”值得\u003C/strong>\u003Cstrong>更深入的\u003C/strong>\u003Cstrong>分析。\u003C/strong>这一条目已\u003Cstrong>明确提及\u003Cspan class=\"\">AI Agent\u003C/span>（智能体）相关的安全威胁\u003C/strong>\u003Cstrong>。\u003C/strong>随着AI Agent越来越多地被用于自动化操作（如自动订票、自动发邮件、自动管理社交账号），Agent被恶意利用来窃取数据、冒充用户、执行未授权操作的风险正在快速上升。Soul作为社交平台对这类风险有切身感知，说明监管方和平台方都已经意识到：智能体安全不是未来的问题，是现在就需要防范的问题。\u003C/section>\n\u003Csection>\n\u003Ch3>电商平台与生活服务：携程、美团、拼多多、得物\u003C/h3>\n\u003C/section>\n\u003Csection>这类平台的AI应用形态不同——携程有”\u003Cspan class=\"\">携程问道\u003C/span>“大模型，美团有AI推荐和AI客服，拼多多有大量AI生成的商品描述和营销内容。它们的合规压力在于：\u003Cstrong>AI内容不仅存在于信息流中，还直接关联交易和服务。\u003C/strong>美团的公告是所有平台中最完整的，完整保留了官方通告中14类问题的全部定义，包括“色情陪聊”、“一键脱衣”、“AI算命”等具体场景。这说明美团的内容治理团队对“AI产品服务和应用程序违规”这一类别做了认真对待——因为这类违规在本地生活服务平台上有直接的商业变现路径（如通过AI算命服务引流、通过AI陪聊变现），风险更为紧迫。\u003C/section>\n\u003Cp> \u003C/p>\n\u003Csection>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"3 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-369 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143454616-300x171.png\" alt=\"\" width=\"300\" height=\"171\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143454616-300x171.png 300w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143454616-1024x582.png 1024w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143454616-768x437.png 768w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143454616.png 1080w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" />\u003C/p>\n\u003Cp style=\"text-align: right;\">从左至右依次为小红书、MiniMax、拼多多\u003C/p>\n\u003Cp style=\"text-align: right;\">图源 ｜ 网信上海 公众号\u003C/p>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection>\n\u003Ch3>开源社区：魔搭社区\u003C/h3>\n\u003C/section>\n\u003Csection>魔搭社区的公告是17家平台中最“特殊”的——它完全不提用户内容，只提模型和数据集的合规审查。这说明\u003Cstrong>开源社区的合规逻辑与内容平台完全不同\u003C/strong>：相较“删帖”和“封禁账号”，它直接采用“下架模型”“撤销上传权限”的整改措施。开源模型的合规审查一旦常态化，将深刻改变开发者的技术获取方式。\u003Cimg loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-370 aligncenter\" src=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143505800-300x205.png\" alt=\"\" width=\"300\" height=\"205\" srcset=\"https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143505800-300x205.png 300w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143505800-768x524.png 768w, https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143505800.png 831w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" />\u003C/p>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"9 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"3 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"box-sizing: border-box;"},"namespaceURI":"http://www.w3.org/1999/xhtml"},"para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\n\u003Cp style=\"text-align: right;\">从左至右依次为2345、魔搭社区\u003C/p>\n\u003Cp style=\"text-align: right;\">图源 ｜ 网信上海 公众号\u003C/p>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection data-pm-slice=\"0 0 []\">\n\u003Csection data-pm-slice=\"3 3 ["para",{"tagName":"section","attributes":{"style":"font-size: 15px;box-sizing: border-box;font-style: normal;font-weight: 400;text-align: justify;color: rgb(62, 62, 62);","data-pm-slice":"0 0 []"},"namespaceURI":"http://www.w3.org/1999/xhtml"}]\">\n\u003Csection>\n\u003Csection>\n\u003Csection>\n\u003Csection>\n\u003Csection>\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection>\n\u003Csection>\n\u003Csection>\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection>\n\u003Csection>\n\u003Csection>\n\u003Csection>\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Ch3>\u003C/h3>\n\u003C/section>\n\u003C/section>\n\u003C/section>\n\u003Csection>\n\u003Ch2>五、结语：4个月专项行动之后，会留下什么？\u003C/h2>\n\u003C/section>\n\u003Csection>从时间线来看，这次专项行动4月底启动，8月底结束。但4个月的集中整治之后，更值得关注的是它会留下什么样的\u003Cstrong>常态化监管框架\u003C/strong>。回顾过往“清朗”系列专项行动的经验，很多整治内容在专项行动结束后被纳入了常态化监管：比如自媒体实名制、算法推荐管理、未成年人网络保护等，都是从”专项行动”起步，最终变成日常监管规则。这次AI专项行动有很高的概率走同样的路径——14类问题清单中的多项内容（大模型备案、AI内容标识、开源模型审查、AI水军治理）很可能在专项行动结束后，被写入或强化为常态化的监管制度。\u003C/p>\n\u003Cp>对于AI行业来说，这意味着\u003Cstrong>合规成本将从“一次性整改”变成“持续性投入”\u003C/strong>。对于普通用户来说，这意味着未来的网络内容环境可能会更透明（AI内容有标识）、更干净（低质AI内容被清理）、更安全（AI水军和虚假信息被打击）。\u003C/p>\n\u003Cp>但一个悬而未决的问题是：“\u003Cstrong>数字泔水”这种带有价值判断的监管概念，如何与平台的内容审核机制对接？\u003C/strong>识别“虚假信息”有技术路径（事实核查、信源验证），但识别”价值空洞”的内容，本质上需要平台对AI生成内容做\u003Cstrong>质量评估\u003C/strong>——这是一个比内容审核更复杂、也更主观的任务。如果这一思路被常态化，平台也将面临前所未有的治理挑战。\u003C/p>\n\u003C/section>\n","https://cdn.lawgenesis.cn/new/posts/2026/05/20260526143102144.jpg","2026.05.26","Blog",[12,14,20,26,32,38,44,50,56,61,67,76,84,90,96,103,110,116,122,128],{"id":4,"category":10,"title":5,"excerpt":6,"date":9,"tag":13,"image":8},"原创转载",{"id":15,"category":10,"title":16,"excerpt":17,"date":18,"tag":13,"image":19},"320","2026年第2期网络主播职业合规能力培训班｜缔零科技受邀授课","缔零科技内容风险审查专家连续两期受邀，站在网络安全学堂的讲台上，为创作者们解读通用版《内容安全具体规范与底线清单》。","2026.05.19","https://cdn.lawgenesis.cn/new/posts/2026/05/20260519134119254.png",{"id":21,"category":10,"title":22,"excerpt":23,"date":24,"tag":13,"image":25},"149","在缔零，他们这样想","这里虽然不发猫\n但是警告：\n本群含猫量极高，但比猫更卷的是这群人的战斗力、活力与能力。听听他们怎么说——","2026.03.06","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310181602777.png",{"id":27,"category":10,"title":28,"excerpt":29,"date":30,"tag":13,"image":31},"142","以认知治理为基石，构建内容安全的未来","在人工智能技术日新月异的今天，我们站在了一个历史性的转折点上。当全球科技巨头竞相追逐算力与参数的极限时，我们选择了一条更为本质、也更具挑战性的道路——致力于成为人类认知治理的基石，在确保AI对齐人类福祉的前提下，推动智能技术的安全发展。这不仅是我们的战略选择，更是深植于组织基因中的使命担当。","2026.03.05","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310080209742.png",{"id":33,"category":10,"title":34,"excerpt":35,"date":36,"tag":13,"image":37},"153","48小时警告Seedance，10亿美元押注OpenAI","2026年2月12日，字节跳动正式发布新一代AI视频生成模型Seedance 2.0。这款被业界称为”地表最强“的AI视频工具，上线当天就在中文互联网掀起狂欢。2月14日，华特迪士尼公司向字节跳动发送了一封措辞严厉的停止侵权函。","2026.02.15","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310182006945.png",{"id":39,"category":10,"title":40,"excerpt":41,"date":42,"tag":13,"image":43},"167","“第五堵墙”——AI技术正在模糊真实与创作的最后一道边界。而对我们来说“完成”，或许比“完美”更重要","创作者未提供的信息，AI却”知道“？\n为何技术展示越惊艳，我们越感到不安？\n“先发布后监管“的模式本身就说明了技术的超前与伦理的滞后\n“完成“或许真的大于“完美“","2026.02.11","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310184045616.png",{"id":45,"category":10,"title":46,"excerpt":47,"date":48,"tag":13,"image":49},"136","内容安全：数字文明的隐形基础设施","在万物互联的时代，内容安全已不再是技术领域的专有名词，而是演变为支撑整个数字文明运转的隐形基础设施。它如同城市地下复杂的管网系统，平时无人察觉，一旦失效，整个社会的信息生态将面临崩溃风险。","2026.02.08","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310061656893-scaled.png",{"id":51,"category":10,"title":52,"excerpt":53,"date":54,"tag":13,"image":55},"226","快手被罚1.19亿：过去的“天花板”，未来的“起步价”","2026年2月6日，北京市网信办对快手开出1.191亿元罚单，在金额上创下同类处罚最新纪录。","2026.02.06","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310221948478.png",{"id":57,"category":10,"title":58,"excerpt":59,"date":54,"tag":13,"image":60},"240","RentAHuman：当AI开始”雇佣”人类，是真风口还是新骗局？","“10万人注册，160万访客，AI正在批量雇佣人类！”\n这条消息在科技圈疯传时，\n一个更小的数字被忽略了：40\n这是RentAHuman平台实际连接的AI Agent数量。\n当10万人类挤在服务器门外，刷新着”登录失败”的页面时，门内只有40个算法”雇主”，发布着”测试链接”和”发推文”的任务——单价从3美元到69美元不等，完成率不到1%\n而故事的讲述者，正是一位用”AI辅助编程”在周末搭起整个网站的加密工程师——他自己，也以69美元/小时的价格，注册成了”可租用人类”。","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310223207623.jpg",{"id":62,"category":10,"title":63,"excerpt":64,"date":65,"tag":13,"image":66},"256","审核员，为什么“人见人嫌”？","后信息时代，平台注意力之争愈演愈烈。\n算法循着你的偏好编织信息茧房，框定了你能看到什么；\n而法律法规与平台规则划下红线，界定了你不能看什么。\n那么，是谁或什么将这份“不能看”的准则落到实处？\n答案是内容审核员。\n是他们的一次次浏览、筛查与判断。","2026.01.15","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310224226530.png",{"id":68,"category":69,"title":70,"excerpt":71,"date":72,"tag":73,"image":74,"externalUrl":75},"133","Media","光谷创业咖啡被投企业缔零科技近日完成新一轮数千万元融资，打造“认知安全”新防线。","近日，缔零科技宣布完成新一轮数千万元融资。本轮融资由中南创投、凯风创投、普朗克创投联合投资，资金将主要用于核心产品“缔零法则”的技术迭代、市场生态构建及认知安全社区的建设。","2025.12.31","外部报道","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310060432732.png","http://www.cctime.com/html/2025-12-15/1725330.htm",{"id":77,"category":78,"title":79,"excerpt":80,"date":81,"tag":73,"image":82,"externalUrl":83},"190","Research","缔零科技荣获“武汉杯”第二届北京大学创新创业大赛银奖","2025年12月17日至18日，“武汉杯”第二届北京大学创新创业大赛决赛颁奖仪式在武汉市中国车谷国际体育文化交流中心隆重举行。","2025.12.25","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310193010979.png","https://mp.weixin.qq.com/s/r3WP4PufYbTomUPT7m2AhQ",{"id":85,"category":69,"title":86,"excerpt":87,"date":81,"tag":73,"image":88,"externalUrl":89},"196","光谷创业咖啡被投企业缔零科技近日完成新一轮数千万元融资。","近日，缔零科技宣布完成新一轮数千万元融资。","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310193448549.png","https://www.toutiao.com/article/7583929576884159022/?&source=m_redirect",{"id":91,"category":10,"title":92,"excerpt":93,"date":94,"tag":13,"image":95},"214","快手有多少次机会避免事故？","违规内容肆虐，是因为审核员不能判断、不想阻拦吗？绝非如此。\n系统不在乎安全。审核不断为增长让步，安全就必然在注册、推荐、审核、决策的每一个环节被牺牲。","2025.12.24","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310210121566.jpg",{"id":97,"category":69,"title":98,"excerpt":99,"date":100,"tag":73,"image":101,"externalUrl":102},"199","缔零科技完成数千万元Pre-A轮融资","18日讯，近日，一站式内容安全解决方案提供商缔零智源科技（深圳）有限公司完成数千万元Pre-A轮融资，本轮投资方为中南创投、凯风创投、普朗克创投。","2025.12.18","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310203136821.png","https://news.qq.com/rain/a/20251218A072OA00",{"id":104,"category":69,"title":105,"excerpt":106,"date":107,"tag":73,"image":108,"externalUrl":109},"209","专注于打造一站式内容安全解决方案，「缔零科技」完成数千万元融资 | 奇绩校友融资新闻","2024 年秋季创业营校友企业、专注于守护人类认知安全的 AI 原生企业「缔零科技」在 11 月完成数千万元天使轮融资，由中南创投、凯风创投、普朗克创投联合投资。所融资金将主要用于其核心产品“缔零法则” 的技术迭代、市场生态构建及认知安全社区建设。","2025.12.15","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310203518366.png","https://mp.weixin.qq.com/s/robktWUGGbjev3oJQzAH7Q",{"id":111,"category":69,"title":112,"excerpt":113,"date":107,"tag":73,"image":114,"externalUrl":115},"203","天使轮数千万元融资，这家公司想成为 AI 时代用户的安全执行顾问","主打认知安全的公司缔零科技最近宣布完成数千万元天使轮融资。","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310203300932.png","https://mp.weixin.qq.com/s/7Tp4E3CbFLhEklYXiI-Z4g",{"id":117,"category":69,"title":118,"excerpt":119,"date":107,"tag":73,"image":120,"externalUrl":121},"206","恭喜！光谷创业咖啡被投企业缔零科技近日完成新一轮数千万元融资，打造“认知安全”新防线","光谷创咖被投企业喜获新一轮融资","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310203425836.png","https://mp.weixin.qq.com/s/qjZXq2vAUqNgYqGcefNGjw",{"id":123,"category":10,"title":124,"excerpt":125,"date":126,"tag":13,"image":127},"183","揭秘GEO灰产：有人正批量给你的AI小助手塞小广告、数据投毒……","我们能做的，就是转发文章，至少让更多人知道：给AI塞答案，正在成为生意。","2025.11.13","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310191111230.jpg",{"id":129,"category":78,"title":130,"excerpt":131,"date":132,"tag":73,"image":133,"externalUrl":134},"193","决赛！AI+智慧治理，独墅湖畔见证创新创业的青春力量","6月15日历经近两月紧张有序的赛程，首届人工智能与智慧治理，大学生创新创业大赛圆满收官，缔零科技荣获二等奖","2025.06.15","https://cdn.lawgenesis.cn/new/posts/2026/03/20260310193142169.webp","https://mp.weixin.qq.com/s/QldgFsb0Ck2ioBxD7_vq5Q",[136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155],{"id":4,"category":10,"title":5,"excerpt":6,"date":9,"tag":13,"image":8},{"id":15,"category":10,"title":16,"excerpt":17,"date":18,"tag":13,"image":19},{"id":21,"category":10,"title":22,"excerpt":23,"date":24,"tag":13,"image":25},{"id":27,"category":10,"title":28,"excerpt":29,"date":30,"tag":13,"image":31},{"id":33,"category":10,"title":34,"excerpt":35,"date":36,"tag":13,"image":37},{"id":39,"category":10,"title":40,"excerpt":41,"date":42,"tag":13,"image":43},{"id":45,"category":10,"title":46,"excerpt":47,"date":48,"tag":13,"image":49},{"id":51,"category":10,"title":52,"excerpt":53,"date":54,"tag":13,"image":55},{"id":57,"category":10,"title":58,"excerpt":59,"date":54,"tag":13,"image":60},{"id":62,"category":10,"title":63,"excerpt":64,"date":65,"tag":13,"image":66},{"id":68,"category":69,"title":70,"excerpt":71,"date":72,"tag":73,"image":74,"externalUrl":75},{"id":77,"category":78,"title":79,"excerpt":80,"date":81,"tag":73,"image":82,"externalUrl":83},{"id":85,"category":69,"title":86,"excerpt":87,"date":81,"tag":73,"image":88,"externalUrl":89},{"id":91,"category":10,"title":92,"excerpt":93,"date":94,"tag":13,"image":95},{"id":97,"category":69,"title":98,"excerpt":99,"date":100,"tag":73,"image":101,"externalUrl":102},{"id":104,"category":69,"title":105,"excerpt":106,"date":107,"tag":73,"image":108,"externalUrl":109},{"id":111,"category":69,"title":112,"excerpt":113,"date":107,"tag":73,"image":114,"externalUrl":115},{"id":117,"category":69,"title":118,"excerpt":119,"date":107,"tag":73,"image":120,"externalUrl":121},{"id":123,"category":10,"title":124,"excerpt":125,"date":126,"tag":13,"image":127},{"id":129,"category":78,"title":130,"excerpt":131,"date":132,"tag":73,"image":133,"externalUrl":134}]