THE SHORT VERSION
- OpenAI, Google, Microsoft and Anthropic back a shared policy token in robots.txt that lets a site block training crawlers for several companies with one entry.OpenAI、谷歌、微软与 Anthropic 支持在 robots.txt 中加入共享 policy token，让站点用一条设置阻止多家公司的训练抓取。
- Anthropic standards lead Kevin Kelley proposed the signal to the Internet Engineering Task Force, and the companies have submitted a draft of the standard.该信号由 Anthropic 标准技术负责人 Kevin Kelley 向 IETF 提出，四家公司已提交标准草案。
- Anthropic has adopted the signal, OpenAI says it will in the coming months and Google and Microsoft plan support next year; Meta welcomed it but did not say it will honour it.Anthropic 已采用，OpenAI 称未来数月落地，谷歌与微软计划明年支持；Meta 表示欢迎，但未说明是否会执行。
One entry instead of many一条记录，而不是很多条
Website operators who want to keep AI developers from training on their content have had to write a separate instruction for each provider. SiliconANGLE reported that a site must add a separate robots.txt entry for every AI company, and if a provider runs several scrapers, each scraper needs its own entry. The file that once told search engines what to index now carries an increasingly long list of AI exceptions.希望阻止 AI 开发者用自家内容训练的网站运营者，过去必须为每一家服务商单独写一条指令。SiliconANGLE 报道称，站点要为每家 AI 公司添加单独的 robots.txt 条目，而如果一家服务商使用多个抓取器，每个抓取器都需要各自的条目。这个原本用于告诉搜索引擎收录范围的文件，如今承载着越来越长的 AI 例外清单。
The proposed standard replaces that list with a single entry. A shared policy token would let a site block all crawlers that collect training data for generative models in one instruction, and the four largest providers would read the same signal. The appeal for a small publisher is obvious: the person maintaining the site no longer has to track which company uses which crawler under which name.拟议的标准用一条记录取代这份清单。共享的 policy token 让站点用一条指令阻止所有为生成式模型收集训练数据的抓取器，而四家最大的服务商读取同一个信号。对小型出版方的好处显而易见：维护站点的人不必再追踪哪家公司用哪个名字的哪个抓取器。
Who proposed it, and where it lives谁提出的，放在哪里
The Verge reported that the shared policy token was proposed by Kevin Kelley, a technical lead for standards at Anthropic, as part of the Internet Engineering Task Force. It would live inside robots.txt, which has long signalled which crawlers may scrape a site for search indexing, archiving or AI training. The companies backing it are OpenAI, Google, Microsoft and Anthropic.The Verge 报道称，共享 policy token 由 Anthropic 标准技术负责人 Kevin Kelley 作为互联网工程任务组（IETF）工作的一部分提出。它将放在 robots.txt 中，这个文件长期用于标明哪些抓取器可以抓取站点内容，用于搜索收录、归档或 AI 训练。支持它的公司是 OpenAI、谷歌、微软与 Anthropic。
The backing matters as much as the mechanism. A robots.txt convention only works if the crawlers honour it, and the companies that run the largest crawlers are the ones proposing the rule. SiliconANGLE reported that the four plan to fold the standard into the Robots Exclusion Protocol and have already submitted a draft version to the IETF, the nonprofit that maintains that protocol. A draft is not a ratified standard, which is the first limit to keep in view.支持者是谁，与机制本身同样重要。robots.txt 约定只有在抓取器遵守时才有效，而运行最大抓取器的公司正是提出这条规则的公司。SiliconANGLE 报道称，四家公司计划把该标准并入 Robots Exclusion Protocol，并已向维护该协议的 IETF 提交草案。草案并不等于已批准的标准，这是首先要看到的限制。
What the setting does not cover这条设置不覆盖什么
The token is about training data, and the draft keeps that narrow. SiliconANGLE reported that the setting would not apply to models that perform exclusively non-generative tasks. It also does not settle the separate question of an AI system fetching a page to answer a user's question, which is the part of crawling that publishers worry about most because it substitutes for a visit.该 token 针对训练数据，草案也把范围限定得很窄。SiliconANGLE 报道称，这条设置不适用于只执行非生成式任务的模型。它也没有解决另一个问题：AI 系统为回答用户问题而抓取页面。这恰恰是出版方最担心的部分，因为它替代了一次访问。
Microsoft addressed that with a separate control called AI Grounding, which lets a site block requests made to answer a user question rather than to collect training data, and it will extend to third-party tools that use Microsoft technology to process prompts. The Verge reported that OpenAI plans similar controls. Microsoft also said sites will still appear in Bing's search results whatever they choose, which keeps search indexing outside the new switch.微软用另一项名为 AI Grounding 的控制来应对：站点可以阻止为回答用户问题而发出的请求，而不是为收集训练数据发出的请求，该功能还将延伸到使用微软技术处理提示词的第三方工具。The Verge 报道称 OpenAI 也计划提供类似控制。微软还表示，无论站点如何选择，它们仍会出现在 Bing 搜索结果中，这把搜索收录留在了新开关之外。
Timelines, and Meta's silence时间表与 Meta 的沉默
The commitments arrive at different speeds. The Verge reported that Anthropic has already adopted the signal, OpenAI says it will implement it in the coming months, and Google and Microsoft plan to support it next year. A publisher who sets the token today would therefore be honoured by one company and ignored, for now, by the others.各家承诺的落地速度不同。The Verge 报道称，Anthropic 已采用该信号，OpenAI 称将在未来数月实施，谷歌与微软计划明年支持。因此，今天设置该 token 的出版方，会被一家公司遵守，而暂时被其他公司忽略。
Meta is the open question. The Verge reported that Meta communications head Andy Stone welcomed the new setting, saying a common signal that conveys preferences to multiple providers at once is an important step, but the company did not say whether it will honour the token. That distinction matters: supporting a standard and complying with it are different commitments, and a publisher reading a press release cannot tell which one Meta has made.Meta 是未解的那个。The Verge 报道称，Meta 传播负责人 Andy Stone 对新设置表示欢迎，称一个能同时向多家服务商传达偏好的共同信号是重要一步，但公司没有说明是否会执行该 token。这一区别很重要：支持标准与遵守标准是两种不同的承诺，而出版方读新闻稿时无法分辨 Meta 给出的是哪一种。
What is established, and what is not哪些已经确定，哪些还没有
Established: OpenAI, Google, Microsoft and Anthropic back a shared robots.txt policy token; Anthropic's standards lead proposed it to the IETF; the four have submitted a draft; Anthropic has adopted it, OpenAI will follow in the coming months and Google and Microsoft next year; Microsoft is adding a separate AI Grounding control; Meta welcomed the idea without committing to honour it. Not established: whether the IETF will adopt the draft, how a site proves a crawler ignored the token, and whether Meta or other providers will comply.已经确定的：OpenAI、谷歌、微软与 Anthropic 支持共享的 robots.txt policy token；Anthropic 的标准负责人已向 IETF 提出；四家公司已提交草案；Anthropic 已采用，OpenAI 未来数月跟进，谷歌与微软明年；微软另加 AI Grounding 控制；Meta 表示欢迎但未承诺执行。尚未确定的：IETF 是否会采纳草案、站点如何证明某个抓取器无视了 token，以及 Meta 或其他服务商是否会遵守。
For readers, the useful test is not whether a company endorses the standard but whether its crawler stops when it reads the token. This site's coverage of Proton's dispute over AI and data access and of Meta's $97.5 billion AI equity raise describes the same tension between platforms and the people whose material they use, and Mistral's decision to open its flagship weights shows a different answer to the same question about who controls the material.对读者而言，有用的检验不是某家公司是否背书该标准，而是它的抓取器读到 token 后是否停下。本站关于Proton 就 AI 与数据访问的争议与Meta 975 亿美元 AI 股权融资的报道，描述了平台与素材所有者之间同样的张力；而Mistral 决定开放旗舰权重则给出了关于“谁控制素材”的另一种回答。
A preference is not a licence偏好不等于授权
A token that tells a crawler to stop is not the same thing as a licence that lets a publisher charge for access. The proposal governs whether training crawlers may fetch a page; it does not set a price, create a contract, or oblige an AI developer to negotiate. Publishers who want payment still need the separate agreements that courts and regulators have been arguing about all year.告诉抓取器停下的标记，与让出版方就访问收费的授权，并不是一回事。这项提案管的是训练抓取器能否抓取页面；它不定价、不构成合同，也不要求 AI 开发者必须谈判。想要收费的出版方，仍需依赖法院与监管机构争论了一整年的那些单独协议。
That gap is why the standard is best read as a floor rather than a settlement. It gives a site operator a cheap, standard way to say no, and it gives the companies a cheap way to be seen to respect that no. Whether the answer changes anything for a newsroom depends on how many crawlers actually stop, which is a question the announcement does not answer.这个缺口正是应当把该标准读作下限而非和解的原因。它给了网站运营者一种廉价、标准的方式说不，也给了各家公司一种廉价的方式被看见在尊重这个不。对新闻编辑室而言，答案是否改变什么，取决于到底有多少抓取器会停下，而这是公告没有回答的问题。
Sources & reporting limits来源与核查范围
Follow Industry to find related stories on your next visit.关注产业动态，下次访问时阅读相关文章。
