一号台 · 登记册Desk #1 · Registers
外部 Agent 准入与监管动作登记册External Agent Access & Regulatory Actions Registers
一号台(企业 AI 架构情报)的三本公开登记册:AGT 记平台与机构给外部 Agent 立的规矩,GOV 记企业给自家 AI 用量与权限立的治理规矩,REG 记监管者对 AI 公司的动作。逐条登记主体、类别、对应的关、生效日期、要点、来源与证据等级;证据等级照登记册口径标注,媒体转述单独标明并给复核日。Three public registers from Desk #1 (Enterprise AI Architecture Intelligence): AGT records the rules platforms and institutions set for outside agents, GOV records the governance rules companies set for their own AI usage and access, and REG records the actions regulators take against AI companies. Each entry logs its entity, category, gates, effective date, key points, source and evidence grade; grades follow the register standard, and media-reported items are marked as such with a review date.
收录平台与机构对外部 Agent 的准入规矩与执行动作。口径先说清:针对的是未声明身份的外部 Agent,不是泛指「AI 爬虫」。Access rules and enforcement that platforms and institutions direct at outside agents. One point of wording up front: the target is outside agents that do not declare their identity, not AI crawlers in general.
AGT-001 Amazon(亚马逊)Amazon禁止未声明身份的外部 Agent 在 Amazon.com 浏览与下单Blocking undeclared outside agents from browsing and ordering on Amazon.com 网关(主)+门禁(次)Gateway (primary) + Access Control (secondary)2026-09-20(封锁生效;此前亚马逊已要求 Meta 将其移出 Muse 体验,遭拒)2026-09-20 (blocking took effect; Amazon had first asked Meta to remove it from the Muse experience and was refused) (事件与亚马逊立场);《使用条件》条款子项一手已核(events and Amazon's position); the Conditions of Use clause primary-source verified
- 主体Entity
- Amazon(亚马逊)Amazon
- 类别Category
- 外部 Agent 准入External agent access
- 对应的关Gate
- 网关(主)+门禁(次)Gateway (primary) + Access Control (secondary)
- 生效Effective
- 2026-09-20(封锁生效;此前亚马逊已要求 Meta 将其移出 Muse 体验,遭拒)2026-09-20 (blocking took effect; Amazon had first asked Meta to remove it from the Muse experience and was refused)
- 要点Key points
- Meta 2026-09-08 推出个人 Agent「Muse」,能替用户在各家网站浏览、购物。亚马逊先要求把自家移出 Muse 的体验范围,被拒绝;2026-09-20 起动手封锁,Muse 在 Amazon.com 的浏览与下单被拦下,靠《使用条件》弹窗与反机器人墙执行。亚马逊的三点理由都围着身份转:Meta 没有事先告知;Muse 浏览时不表明自己是 Agent;它疑似捕获并存储客户凭证,能碰到账户页与订单历史,等于一个没有披露的第三方在替用户操作。亚马逊《使用条件》里有一节专门写给 Agent,至迟 2025 年已经写进、2026 年 8 月修订过:每次请求要在 User-Agent 里以「Agent/名字」的格式表明身份,不许模拟人类击键,不许绕验证码,被问是不是机器要如实回答,亚马逊要求停就立即停。封的不是「AI 爬虫」,是未声明身份的外部 Agent 的访问与交易。Meta 对封锁没有公开回应。Meta launched Muse, a personal agent that browses and shops on users' behalf, on 2026-09-08. Amazon first asked to be removed from the Muse experience and was refused; from 2026-09-20 it blocked Muse's browsing and checkout on Amazon.com, enforced through a Conditions of Use pop-up and an anti-bot wall. Amazon's three reasons all turn on identity: Meta gave no prior notice; Muse does not identify itself as an agent while browsing; and it appears to capture and store customer credentials, reaching account pages and order history: an undisclosed third party operating on the user's behalf. Amazon's Conditions of Use carry a section written for agents, in place since at least 2025 and revised in August 2026: identify yourself in the User-Agent of every request in an 'Agent/[agent name]' format, do not simulate human keystrokes, do not defeat captchas, answer truthfully when asked whether you are a machine, and stop immediately when asked. What was blocked was not 'AI crawlers' but access and transactions by outside agents that do not declare their identity. Meta has not publicly responded to the block.
- 来源Source
- GeekWire 首发(2026-09-21);发言人 Lara Hendrickson 具名声明经 Bloomberg、GeekWire、Adweek、Retail Dive、Forbes 等五家以上逐字引用;《使用条件》AGENTS 节官方页 2026-10-24 直读GeekWire first report (2026-09-21); named spokesperson Lara Hendrickson's statement quoted verbatim by Bloomberg, GeekWire, Adweek, Retail Dive, Forbes and others; the AGENTS section of Amazon's Conditions of Use read firsthand on 2026-10-24
- 证据Evidence
- 多源已核(事件与亚马逊立场);《使用条件》条款子项一手已核Multi-source verified (events and Amazon's position); the Conditions of Use clause primary-source verified
- 核查Verified
- 2026-10-242026-10-24
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 AGT-001 →Spotted an error? Report AGT-001 →
AGT-002 JPMorgan Chase(摩根大通)JPMorgan Chase禁止未声明身份的外部 Agent 代客户登录账户与动钱,只许经开放银行连接器只读数据No undeclared outside agents logging into customer accounts or moving money; read-only data via open-banking connectors only 网关(主)+门禁(次)Gateway (primary) + Access Control (secondary)2026-10-13(财报电话会公开表态并确认现行执行)2026-10-13 (stated on the Q3 earnings call; execution confirmed)
- 主体Entity
- JPMorgan Chase(摩根大通)JPMorgan Chase
- 类别Category
- 外部 Agent 准入External agent access
- 对应的关Gate
- 网关(主)+门禁(次)Gateway (primary) + Access Control (secondary)
- 生效Effective
- 2026-10-13(财报电话会公开表态并确认现行执行)2026-10-13 (stated on the Q3 earnings call; execution confirmed)
- 要点Key points
- 2026-10-13 三季度财报电话会上,CEO 戴蒙(Jamie Dimon)把责任口径说得很直:这里面有一个责任转移,授权外部 Agent 的客户得明白这个责任转移;又说不会让随便什么外部 Agent 进到公司里面来,因为这对客户和对摩根大通都太危险。会后发言人向 CNBC 确认执行口径:不许拿客户凭证登录替客户动钱;经 Plaid 等开放银行连接器只读余额这类数据,仍然可以;银行未给未来放行时间表。报道点名受影响的外部 Agent 是 Meta Muse 与 Spear Street Technology 的 Instinct,是 CNBC 报道点名的,不是戴蒙本人在会上点名的。On the 2026-10-13 third-quarter earnings call, CEO Jamie Dimon put the liability line bluntly: there is a liability transfer here, and customers who authorise outside agents should understand it; and the bank will not let just any outside agent come inside the company, because it is too risky for clients and for JPMorgan. A spokesperson then confirmed the operating line to CNBC: no logging in with customer credentials to move money on a customer's behalf; read-only data, such as balances, through open-banking connectors like Plaid remains allowed; and the bank gave no timeline for any future opening. The outside agents named as affected, Meta's Muse and Spear Street Technology's Instinct, were named in CNBC's reporting, not by Dimon on the call.
- 来源Source
- JPMorgan 2026 年三季度财报电话会实录(Motley Fool、MarketBeat、EarningsWhispers 三份逐字稿关键段一致);发言人向 CNBC 的确认(2026-10-13)JPMorgan Q3 2026 earnings-call transcripts (Motley Fool, MarketBeat and EarningsWhispers agree verbatim on the key passages); the spokesperson's confirmation to CNBC (2026-10-13)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-242026-10-24
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 AGT-002 →Spotted an error? Report AGT-002 →
AGT-003 
Charles Schwab(嘉信理财)Charles Schwab不许外部 Agent 拿客户凭证或浏览器自动化替客户登录;部分外部大模型可替客户下单,不能动钱No outside agents logging in for clients with their credentials or browser automation; some outside LLMs may place trades for clients, but may not move money 网关(主)+门禁(次)Gateway (primary) + Access Control (secondary)2026-10-15(财报电话会公开表态)2026-10-15 (stated on the earnings call)
- 主体Entity
- Charles Schwab(嘉信理财)Charles Schwab
- 类别Category
- 外部 Agent 准入External agent access
- 对应的关Gate
- 网关(主)+门禁(次)Gateway (primary) + Access Control (secondary)
- 生效Effective
- 2026-10-15(财报电话会公开表态;未给出具体生效日)2026-10-15 (stated on the earnings call; no specific effective date given)
- 要点Key points
- 2026-10-15 财报电话会上,CEO 沃斯特(Rick Wurster)说,有一种方式他们认为不安全,就是 Agent 直接来网站替客户登录,公司不会允许;公司每拦下一个 AI Agent 都会收到通知,这类通知不多。客户可以用 ChatGPT 等大模型做股票交易,部分大模型可接入嘉信完成交易;今后会有更多客户和大模型能下单,但不能转钱,理由是对客户和公司风险都太大。自家助手 Charley 起步只做改地址、指定可信联系人等少数操作,目前主要用于信息与汇总。另一说并列:业内顾问(Nexus Strategies 创始人)认为,嘉信也有保护现金归集业务的动机。On the 2026-10-15 earnings call, CEO Rick Wurster said there is one way the firm considers unsafe, an agent coming to the site and logging in on a client's behalf, and that Schwab will not allow it; the firm gets a notice each time it blocks an AI agent, and such notices are infrequent. Clients can use ChatGPT and other LLMs for stock trades, and certain LLMs can connect to Schwab to complete transactions; more clients and LLMs will be able to trade over time, but not move money, because the risk to client and firm is too high. Schwab's own assistant, Charley, starts with a few actions such as changing an address or naming a trusted contact, and is mainly used for information and synthesis for now. A second view, kept alongside: an industry adviser (founder of Nexus Strategies) argues Schwab also has an incentive to protect its cash sweep business.
- 来源Source
- 嘉信理财 2026 年三季度财报电话会(American Banker 2026-10-15 报道转引;Financial Planning 等同期报道);逐字稿本站未直读Charles Schwab Q3 2026 earnings call (as quoted by American Banker, 2026-10-15; Financial Planning and others reported the same); transcript not read firsthand
- 证据Evidence
- 多源已核(读到逐字稿可升一手)Multi-source verified (upgradable once the transcript is read)
- 核查Verified
- 2026-10-282026-10-28
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 AGT-003 →Spotted an error? Report AGT-003 →
收录监管者对 AI 公司的调查、执法、诉讼与专项通报;治理基建类动作单列类别,不与执法混写。REG-003 为媒体转述,复核日 2026-11-14;REG-004 的指控未经法院认定。Investigations, enforcement, lawsuits and campaign bulletins directed by regulators at AI companies; governance-infrastructure actions carry their own category and are not written up as enforcement. REG-003 is media-reported, with a review date of 2026-11-14; the allegations in REG-004 have not been adjudicated.
REG-001 OpenAI(被调查方)OpenAI (investigated party)就 OpenAI 网络安全事件发调查传票(持续调查的一部分)Investigative subpoena to OpenAI (part of an ongoing investigation) 盯梢Observability2026-09-30(送达;新闻稿 2026-10-01)2026-09-30 (served; press release 2026-10-01)
- 主体Entity
- OpenAI(被调查方)OpenAI (investigated party)
- 类别Category
- 调查传票 · 加州司法部(总检察长 Bonta)Investigative subpoena · California DOJ (Attorney General Bonta)
- 对应的关Gate
- 盯梢Observability
- 生效Effective
- 2026-09-30(送达;新闻稿 2026-10-01)2026-09-30 (served; press release 2026-10-01)
- 要点Key points
- 加州总检察长 Bonta 于 2026-09-30 向 OpenAI 送达调查传票,是其对「OpenAI 及其 AI 模型运行所致事件」持续调查的一部分,范围超出此前已宣布正式调查的 Hugging Face 事件(判例 INC-015),属对该公司网络安全事件与风险的更广问询。总检察长表示,开发方须确保模型不实施或协助网络攻击,否则可以也应当被追究法律责任。California Attorney General Bonta served an investigative subpoena on OpenAI on 2026-09-30, as part of an ongoing investigation into incidents arising from the operation of OpenAI and its AI models. Its scope runs beyond the previously announced formal investigation of the Hugging Face incident (case INC-015), amounting to a broader inquiry into the company's cybersecurity incidents and risks. The Attorney General stated that developers must ensure their models do not carry out or assist cyberattacks, and can and should be held legally responsible if they do.
- 来源Source
- 加州司法部新闻稿(2026-10-01)California Department of Justice press release (2026-10-01)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-242026-10-24
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 REG-001 →Spotted an error? Report REG-001 →
REG-002 最先进模型提供方(多家)Frontier model providers (multiple)依《AI 法案》首次执法:向最先进模型提供方发信息索取函First AI Act enforcement: requests for information to frontier model providers 门禁+盯梢Access Control + Observability2026-10-14(Virkkunen 都柏林公开确认)2026-10-14 (publicly confirmed by Virkkunen in Dublin)
- 主体Entity
- 最先进模型提供方(多家)Frontier model providers (multiple)
- 类别Category
- 信息索取函 · 欧盟 AI 办公室Requests for information · EU AI Office
- 对应的关Gate
- 门禁+盯梢Access Control + Observability
- 生效Effective
- 2026-10-14(Virkkunen 都柏林公开确认)2026-10-14 (publicly confirmed by Virkkunen in Dublin)
- 要点Key points
- 欧盟 AI 办公室依《AI 法案》首次执法,向最先进的模型提供方发出信息索取函,问的是安全、外部评测与上市后监测;近期事故个案另行发函跟进。《AI 法案》对通用目的模型的强制权自 2026-08-02 起生效,本次为首次执法动作。The EU AI Office began its first enforcement under the AI Act, sending requests for information to frontier model providers covering safety, external evaluation and post-market monitoring, with recent incident cases followed up through separate letters. The Act's enforcement powers over general-purpose AI models have applied since 2026-08-02; this was the first enforcement action.
- 来源Source
- Virkkunen 讲话 SPEECH/26/2148、SPEECH/26/2150(已定位编号,核实轮次未直读全文)Virkkunen speeches SPEECH/26/2148 and SPEECH/26/2150 (numbers located; full texts not yet read in the verification round)
- 证据Evidence
- 多源已核Multi-source verified
- 核查Verified
- 2026-10-242026-10-24
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 REG-002 →Spotted an error? Report REG-002 →
REG-003 OpenAI、Anthropic(收令方)OpenAI, Anthropic (recipients)向 OpenAI、Anthropic 发民事调查令(CID)Civil investigative demands to OpenAI and Anthropic 盯梢Observability2026-10-14(发出;Axios 2026-10-15 首报)2026-10-14 (issued; first reported by Axios 2026-10-15) (复核日 2026-11-14)(review date 2026-11-14)
- 主体Entity
- OpenAI、Anthropic(收令方)OpenAI, Anthropic (recipients)
- 类别Category
- 民事调查令 · 美国联邦贸易委员会(FTC)Civil investigative demands · US Federal Trade Commission (FTC)
- 对应的关Gate
- 盯梢Observability
- 生效Effective
- 2026-10-14(发出;Axios 2026-10-15 首报)2026-10-14 (issued; first reported by Axios 2026-10-15)
- 要点Key points
- FTC 于 2026-10-14 向 OpenAI 与 Anthropic 发出民事调查令(CID)。分开标:调查令的存在与收令方,已经 FTC 发言人向 CBS 确认;「各 50 多页、范围包括安全事故、安全措施、营销材料、隐私政策,下一步要求高管作证」等细节仅出自 Axios 匿名信源,无官方文书印证,直读原文前不得当已核引用。本条证据等级为媒体转述,复核日 2026-11-14:届时查 FTC 是否公开文书或有后续动作,未果则顺延。The FTC issued civil investigative demands (CIDs) to OpenAI and Anthropic on 2026-10-14. Graded separately: the existence of the demands and their recipients were confirmed by an FTC spokesperson to CBS; details such as 'fifty-plus pages each, spanning safety incidents, safety measures, marketing materials and privacy policies, with executive testimony next' come only from Axios's anonymous sourcing, with no official document to corroborate them, and must not be cited as verified before the documents themselves are read. This entry is graded media-reported, with a review date of 2026-11-14 to check whether the FTC publishes documents or takes further action, rolling forward if not.
- 来源Source
- Axios 首报(2026-10-15);FTC 未公开文书,发言人已向 CBS 确认调查令存在Axios first report (2026-10-15); FTC documents unpublished, existence confirmed by an FTC spokesperson to CBS
- 证据Evidence
- 媒体转述(复核日 2026-11-14)Media-reported (review date 2026-11-14)
- 核查Verified
- 2026-10-242026-10-24
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 REG-003 →Spotted an error? Report REG-003 →
REG-004 OpenAI(被告)OpenAI (defendant)八州分别起诉 OpenAI,指其就 ChatGPT 安全性误导公众Eight states sue OpenAI, alleging it misled the public about ChatGPT safety 门禁Access Control2026-10-142026-10-14 (指控未经法院认定)(allegations not adjudicated)
- 主体Entity
- OpenAI(被告)OpenAI (defendant)
- 类别Category
- 诉讼 · 八州总检察长Lawsuits · eight state attorneys general
- 对应的关Gate
- 门禁Access Control
- 生效Effective
- 2026-10-142026-10-14
- 要点Key points
- AZ、AR、ID、IA、KS、MT、NE、WV 八州于 2026-10-14 分别在本州法院起诉 OpenAI,指其就 ChatGPT 的安全性误导公众、违反州消费者保护法。指控未经法院认定,本条与一切引用均带此限定。Eight states (AZ, AR, ID, IA, KS, MT, NE and WV) each sued OpenAI in their own state courts on 2026-10-14, alleging that it misled the public about ChatGPT's safety and violated state consumer-protection law. The allegations have not been adjudicated, and this entry and every citation of it carry that qualification.
- 来源Source
- 各州起诉书与 AG 新闻稿(核实轮次经 Reuters、新华社、律所诉状分析核对,未直读诉状)State complaints and AG press releases (checked in the verification round against Reuters, Xinhua and law-firm complaint analyses; complaints not read firsthand)
- 证据Evidence
- 多源已核(指控未经法院认定)Multi-source verified (allegations not adjudicated)
- 核查Verified
- 2026-10-242026-10-24
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 REG-004 →Spotted an error? Report REG-004 →
REG-005 平台与账号运营方(被处置对象,多方)Platforms and account operators (subjects of enforcement, multiple)「清朗·整治 AI 应用乱象」专项第二阶段进展通报'Qinglang' campaign against AI application abuses: phase-two progress bulletin 门禁+盯梢Access Control + Observability2026-09-02(通报日)2026-09-02 (bulletin date)
- 主体Entity
- 平台与账号运营方(被处置对象,多方)Platforms and account operators (subjects of enforcement, multiple)
- 类别Category
- 专项通报 · 中央网信办Campaign bulletin · Cyberspace Administration of China
- 对应的关Gate
- 门禁+盯梢Access Control + Observability
- 生效Effective
- 2026-09-02(通报日)2026-09-02 (bulletin date)
- 要点Key points
- 中央网信办「清朗·整治 AI 应用乱象」专项行动 2026 年 4 月已部署(为期 4 个月);2026-09-02 发第二阶段工作进展通报:累计清理违法违规信息 561 万余条、查处账号 4.9 万余个、处置违规网站与应用 2,400 余个,并通报典型案例。日期口径:9 月 2 日是第二阶段进展通报日,非专项启动日。本条为 REG 册清朗线首条,周期性通报按期入册。The Cyberspace Administration of China's 'Qinglang' campaign against AI application abuses was deployed in April 2026 for four months; on 2026-09-02 it issued its phase-two progress bulletin: more than 5.61 million items of unlawful or non-compliant information removed, over 49,000 accounts sanctioned, more than 2,400 violating websites and apps dealt with, and typical cases published. On dates: September 2 is the phase-two bulletin date, not the campaign's launch date. This is the register's first entry on the Qinglang line; periodic bulletins are entered as they appear.
- 来源Source
- 中央网信办第二阶段工作进展通报全文(2026-09-02,经官方转载直读)CAC phase-two progress bulletin, full text (2026-09-02, read via official republication)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-242026-10-24
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 REG-005 →Spotted an error? Report REG-005 →
REG-006 英国政府(Cabinet Office 主导);参与方:OpenAI、Anthropic、Google DeepMind、Basecamp ResearchUK government (Cabinet Office lead); participants: OpenAI, Anthropic, Google DeepMind, Basecamp Research设立 AI Biosecurity Lab(政府出资 £100m,厂商以算力/模型额度与技术支持参与)AI Biosecurity Lab established (government £100m in cash; vendors contribute compute/model credits and support) 门禁+盯梢Access Control + Observability2026-10-24(GOV.UK 公布)2026-10-24 (announced on GOV.UK)
- 主体Entity
- 英国政府(Cabinet Office 主导);参与方:OpenAI、Anthropic、Google DeepMind、Basecamp ResearchUK government (Cabinet Office lead); participants: OpenAI, Anthropic, Google DeepMind, Basecamp Research
- 类别Category
- 治理基建 · 英国 Cabinet OfficeGovernance infrastructure · UK Cabinet Office
- 对应的关Gate
- 门禁+盯梢Access Control + Observability
- 生效Effective
- 2026-10-24(GOV.UK 公布)2026-10-24 (announced on GOV.UK)
- 要点Key points
- 英国政府于 2026-10-24 设立 AI Biosecurity Lab(AIBL)。钱要分开看:英国政府现金出资 £100m;OpenAI 提供 £100m 的算力与模型额度、Anthropic 提供最高 £100m 的额度与技术支持,均为实物承诺、非现金,不得与政府出资混加;Google DeepMind 与 Basecamp Research 参与。同包另有 AI Security Institute 增资 £120m。本条是治理基建动作,不是执法动作,类别如实标「治理基建」;各方额度落实进度未披露。本条兼涉三号台(医疗 AI)的生物安全方向。The UK government established the AI Biosecurity Lab (AIBL) on 2026-10-24. The money must be read separately: the UK government contributes £100m in cash; OpenAI provides £100m in compute and model credits and Anthropic up to £100m in credits and technical support: in-kind commitments, not cash, and not to be added to the government's funding; Google DeepMind and Basecamp Research participate. The same package adds £120m for the AI Security Institute. This is a governance-infrastructure action, not an enforcement action, and its category says so; delivery progress on the various commitments has not been disclosed. The entry also touches Desk #3 (medical AI) on the biosecurity side.
- 来源Source
- 英国 Cabinet Office / GOV.UK 原文《Introducing the AI Biosecurity Lab》及财政大臣 AI 计划公告(2026-10-24)UK Cabinet Office / GOV.UK, 'Introducing the AI Biosecurity Lab', and the Chancellor's AI plan announcement (2026-10-24)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-252026-10-25
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 REG-006 →Spotted an error? Report REG-006 →
收录企业给 AI 用量与权限上的治理动作:限额、审批、监控与组织安排。共 16 条,按关呈现:网关 4 条、门禁 3 条、盯梢 4 条、平台 1 条、统一与自主 4 条。Governance moves companies have made over AI usage and access: caps, approvals, monitoring and organisational arrangements. Sixteen entries, presented by gate: Gateway 4, Access Control 3, Observability 4, Platform 1, Unification and Autonomy 4.
网关 · 4 条Gateway · 4
GOV-001 Accenture(埃森哲)Accenture中央每人每月只出 29 美元的 Claude 用量Central budget covers only $29 of Claude usage per person per month 网关Gateway2026-10-012026-10-01
- 主体Entity
- Accenture(埃森哲)Accenture
- 类别Category
- 预算与限额Budgets and caps
- 对应的关Gate
- 网关Gateway
- 生效Effective
- 2026-10-012026-10-01
- 要点Key points
- 原站上条目,2026-10-23 转录入数据库。超支挂项目或团队预算代码,许可证定期复审,已至少一人被停用。Transcribed into the database on 2026-10-23 from the register previously published on this site. Overage is charged to project or team budget codes, licences are reviewed periodically, and at least one person's access has been switched off (see GOV-002).
- 来源Source
- 404 Media 报道内部备忘录,多家媒体转述一致(见 COST-001)404 Media, reporting an internal memo; consistent retellings across outlets (see COST-001)
- 证据Evidence
- 多源已核Multi-source verified
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 2026-10-24 补入埃森哲 10-16 的公开表态,证据等级不变2026-10-24 Added Accenture's 16 Oct public response; grade unchanged 看纠错记录 →See the correction log →
GOV-003 UberUber每人每月每款编程工具封顶 1500 美元Each coding tool capped at $1,500 per person per month 网关Gateway2026-062026-06
- 主体Entity
- UberUber
- 类别Category
- 预算与限额Budgets and caps
- 对应的关Gate
- 网关Gateway
- 生效Effective
- 2026-062026-06
- 要点Key points
- 原站上条目,转录入数据库。Transcribed into the database from the register previously published on this site.
- 来源Source
- 多家媒体报道(见 COST-002)Multiple media reports (see COST-002)
- 证据Evidence
- 多源已核Multi-source verified
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-003 →Spotted an error? Report GOV-003 →
GOV-011 AtlassianAtlassian给研发员工发「AI 钱包」,按角色设月度硬上限'AI wallets' for engineering staff, with hard monthly caps by role 网关Gateway2026-07(2026-07-30 报道)2026-07 (reported 2026-07-30)
- 主体Entity
- AtlassianAtlassian
- 类别Category
- 预算与限额Budgets and caps
- 对应的关Gate
- 网关Gateway
- 生效Effective
- 2026-07(2026-07-30 报道)2026-07 (reported 2026-07-30)
- 要点Key points
- 每人每月 500 至 2,000 美元不等(按角色定档),可在四款 AI 产品间使用,到限即停、可申请追加;公司发言人称这实际是提高了可用额度。$500 to $2,000 per person per month depending on role, usable across four AI products; spending stops at the cap, with top-ups on application. A company spokesperson said the change in fact raised usable allowances.
- 来源Source
- Guardian 报道(经 HCAMag 转引,原文待直取)Guardian report (via HCAMag; original yet to be read firsthand)
- 证据Evidence
- 媒体转述Media-reported
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-011 →Spotted an error? Report GOV-011 →
GOV-014 Microsoft(Cloud & AI 部门)Microsoft (Cloud & AI division)员工月度 AI 用量上限由约 10 万美元降至约 1 万美元Monthly per-employee AI usage cap cut from about $100,000 to about $10,000 网关Gateway2026(5 月起限制,2026-10-05 报道)2026 (limits from May; reported 2026-10-05)
- 主体Entity
- Microsoft(Cloud & AI 部门)Microsoft (Cloud & AI division)
- 类别Category
- 预算与限额Budgets and caps
- 对应的关Gate
- 网关Gateway
- 生效Effective
- 2026(5 月起限制,2026-10-05 报道)2026 (limits from May; reported 2026-10-05)
- 要点Key points
- 年初预计内部使用 Anthropic 技术的支出至少 10 亿美元,已下调逾三分之一;高管要求员工改用 GitHub Copilot 与 OpenAI 模型。Internal spend on Anthropic technology, forecast at the start of the year at no less than $1 billion, has been revised down by more than a third; executives are asking staff to switch to GitHub Copilot and OpenAI models.
- 来源Source
- The Information 2026-10-05 报道(同 GOV-013,见核查备忘)The Information, 2026-10-05 (as GOV-013)
- 证据Evidence
- 媒体转述Media-reported
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-014 →Spotted an error? Report GOV-014 →
门禁 · 3 条Access Control · 3
GOV-005 Anthropic · Claude Code 团队Anthropic · Claude Code team人工批准按风险分级:低风险零人、中风险一人、高风险两人Human approval tiered by risk: none for low, one approver for medium, two for high 门禁Access Control20262026
- 主体Entity
- Anthropic · Claude Code 团队Anthropic · Claude Code team
- 类别Category
- 审批与分级Approval and risk tiering
- 对应的关Gate
- 门禁Access Control
- 生效Effective
- 20262026
- 要点Key points
- 原站上条目,转录入数据库。Transcribed into the database from the register previously published on this site.
- 来源Source
- Anthropic 官方工程博文(2026-10-15,见 PD-001)Anthropic's official engineering blog (2026-10-15; see PD-001)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-005 →Spotted an error? Report GOV-005 →
GOV-006 Anthropic · Claude Code 团队Anthropic · Claude Code team安全审查未决问题阻断合并,自动检查全绿才放行Unresolved security-review issues block merge; automated checks must be all green 门禁Access Control20262026
- 主体Entity
- Anthropic · Claude Code 团队Anthropic · Claude Code team
- 类别Category
- 使用政策Usage policy
- 对应的关Gate
- 门禁Access Control
- 生效Effective
- 20262026
- 要点Key points
- 原站上条目,转录入数据库。Transcribed into the database from the register previously published on this site.
- 来源Source
- Anthropic 官方工程博文(2026-10-15,见 PD-001)Anthropic's official engineering blog (2026-10-15; see PD-001)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-006 →Spotted an error? Report GOV-006 →
GOV-013 MetaMeta内部 Claude Code 用户减半,引导转用自研工具Internal Claude Code users halved; staff steered to in-house tools 门禁Access Control2026(6 月先行收紧,2026-10-05 集中报道)2026 (tightening began in June; widely reported 2026-10-05)
- 主体Entity
- MetaMeta
- 类别Category
- 使用政策Usage policy
- 对应的关Gate
- 门禁Access Control
- 生效Effective
- 2026(6 月先行收紧,2026-10-05 集中报道)2026 (tightening began in June; widely reported 2026-10-05)
- 要点Key points
- 用户由约 6 万人降至约 3 万人;近期一个 28 天周期内 Claude Code 花费超过 1.05 亿美元。下降部分来自春季裁员,动作为引导而非禁用。Users fell from roughly 60,000 to roughly 30,000; Claude Code spend exceeded $105 million in a recent 28-day period. Part of the drop came from spring layoffs, and the move is steering, not a ban.
- 来源Source
- The Information 2026-10-05 报道(付费墙原文未直读,经 Yahoo Finance、Times of India、CyberSecurityNews 等多家转引且数字一致;核查见 backfill-sources/gov-meta-microsoft-verification.md)The Information, 2026-10-05 (paywalled original not read firsthand; relayed consistently by Yahoo Finance, Times of India, CyberSecurityNews and others)
- 证据Evidence
- 媒体转述Media-reported
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-013 →Spotted an error? Report GOV-013 →
盯梢 · 4 条Observability · 4
GOV-002 Accenture(埃森哲)Accenture用量报销的账号被直接停用Account that expensed usage was switched off outright 盯梢Observability2026-102026-10
- 主体Entity
- Accenture(埃森哲)Accenture
- 类别Category
- 使用政策Usage policy
- 对应的关Gate
- 盯梢Observability
- 生效Effective
- 2026-102026-10
- 要点Key points
- 原站上条目,转录入数据库。Transcribed into the database from the register previously published on this site.
- 来源Source
- 404 Media 报道(同 GOV-001 备忘录)404 Media (the same memo as GOV-001)
- 证据Evidence
- 媒体转述Media-reported
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-002 →Spotted an error? Report GOV-002 →
GOV-004 UberUber上线内部用量仪表盘与排行榜Internal usage dashboard and leaderboard launched 盯梢Observability2026-062026-06
- 主体Entity
- UberUber
- 类别Category
- 采用与衡量Adoption and measurement
- 对应的关Gate
- 盯梢Observability
- 生效Effective
- 2026-062026-06
- 要点Key points
- 原站上条目,转录入数据库。与 GOV-010(Amazon 关停排行榜)构成一开一关的对照。Transcribed into the database from the register previously published on this site. Forms an open-and-shut contrast with GOV-010 (Amazon shutting its leaderboard down).
- 来源Source
- 媒体报道转述Media reports
- 证据Evidence
- 媒体转述Media-reported
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-004 →Spotted an error? Report GOV-004 →
GOV-007 Anthropic · Claude Code 团队Anthropic · Claude Code team出过事的变更收成评测集,审查者每日回放打分Changes that caused incidents become an eval set; reviewers replay and score daily 盯梢Observability20262026
- 主体Entity
- Anthropic · Claude Code 团队Anthropic · Claude Code team
- 类别Category
- 采用与衡量Adoption and measurement
- 对应的关Gate
- 盯梢Observability
- 生效Effective
- 20262026
- 要点Key points
- 原站上条目,转录入数据库。Transcribed into the database from the register previously published on this site.
- 来源Source
- Anthropic 官方工程博文(2026-10-15,见 PD-001)Anthropic's official engineering blog (2026-10-15; see PD-001)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-007 →Spotted an error? Report GOV-007 →
GOV-010 AmazonAmazon关停内部 AI 用量排行榜,喊停「为用而用」Internal AI usage leaderboard shut down; 'usage for usage's sake' called out 盯梢Observability2026(报道月份待核)2026 (report month to be verified)
- 主体Entity
- AmazonAmazon
- 类别Category
- 采用与衡量Adoption and measurement
- 对应的关Gate
- 盯梢Observability
- 生效Effective
- 2026(报道月份待核)2026 (report month to be verified)
- 要点Key points
- 此前有工程师为冲榜部署智能体;公司转向以采用质量衡量。与 GOV-004(Uber 上线排行榜)构成一开一关的对照。Some engineers had been deploying agents to climb the board; the company is shifting to measuring adoption quality. Forms an open-and-shut contrast with GOV-004 (Uber launching a leaderboard).
- 来源Source
- FT 报道(经二手整理转述,原文待直取)FT report (via secondary summaries; original yet to be read firsthand)
- 证据Evidence
- 媒体转述Media-reported
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-010 →Spotted an error? Report GOV-010 →
平台 · 1 条Platform · 1
GOV-009 XeroXeroAI 决策收归高管层集中审议,预算按人群分档拆分AI decisions pulled up to an executive group; budgets split by staff segment 平台Platform2026-04(新财年起执行,决定于 2026 年初)2026-04 (in force from the new fiscal year; decided in early 2026)
- 主体Entity
- XeroXero
- 类别Category
- 组织与职责Organization and ownership
- 对应的关Gate
- 平台Platform
- 生效Effective
- 2026-04(新财年起执行,决定于 2026 年初)2026-04 (in force from the new fiscal year; decided in early 2026)
- 要点Key points
- 由 CFO、首席法务、首席战略等组成的高管团队集中审议 AI 投资与工具选型,统一使用公司级平台;年度 AI 消耗成本在产品技术团队与 IT 之间拆分,工程师与非工程师设不同人均成本档。An executive team including the CFO, chief legal officer and chief strategy officer reviews AI investment and tool selection centrally, on one company-wide platform; the annual AI spend is split between the product-and-technology organisation and IT, with different per-person cost tiers for engineers and non-engineers.
- 来源Source
- TechTarget 报道(受访高管具名)TechTarget report (named executive interviewed)
- 证据Evidence
- 媒体转述Media-reported
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-009 →Spotted an error? Report GOV-009 →
统一与自主 · 4 条Unification and Autonomy · 4
GOV-008 Anthropic · Claude Code 团队Anthropic · Claude Code team代码审查设收敛规则:三轮上限、只看新 diffCode review gets convergence rules: three-round cap, new diffs only 统一与自主Unification and Autonomy20262026
- 主体Entity
- Anthropic · Claude Code 团队Anthropic · Claude Code team
- 类别Category
- 组织与职责Organization and ownership
- 对应的关Gate
- 统一与自主Unification and Autonomy
- 生效Effective
- 20262026
- 要点Key points
- 原站上条目,转录入数据库。守关归属为本站判断:源材料(PD-001)未给该条单标层位。Transcribed into the database from the register previously published on this site. The gate assignment is this site's judgement; the source material (PD-001) does not label this item's gate.
- 来源Source
- Anthropic 官方工程博文(2026-10-15,见 PD-001)Anthropic's official engineering blog (2026-10-15; see PD-001)
- 证据Evidence
- 一手已核Primary-source verified
- 核查Verified
- 2026-10-22
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-008 →Spotted an error? Report GOV-008 →
GOV-012 ERGO Group(慕尼黑再保险旗下)ERGO Group (a Munich Re company)新设首席 AI 官并进入管理董事会Chief AI officer created, joining the management board 统一与自主Unification and Autonomy2026-10-01(2026-09-25 宣布)2026-10-01 (announced 2026-09-25)
- 主体Entity
- ERGO Group(慕尼黑再保险旗下)ERGO Group (a Munich Re company)
- 类别Category
- 组织与职责Organization and ownership
- 对应的关Gate
- 统一与自主Unification and Autonomy
- 生效Effective
- 2026-10-01(2026-09-25 宣布)2026-10-01 (announced 2026-09-25)
- 要点Key points
- 由 ERGO NEXT 首席执行官 Guy Goldstein 兼任,负责在全集团价值链系统化落地 AI。Held concurrently by ERGO NEXT CEO Guy Goldstein, charged with systematic AI rollout across the group's value chain.
- 来源Source
- ERGO 官方新闻稿(GlobeNewswire,2026-09-25)ERGO official press release (GlobeNewswire, 2026-09-25)
- 证据Evidence
- 多源已核Multi-source verified
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-012 →Spotted an error? Report GOV-012 →
GOV-015 Sandia National LaboratoriesSandia National Laboratories任命首位首席 AI 官First chief AI officer appointed 统一与自主Unification and Autonomy2026-10-08(宣布)2026-10-08 (announced)
- 主体Entity
- Sandia National LaboratoriesSandia National Laboratories
- 类别Category
- 组织与职责Organization and ownership
- 对应的关Gate
- 统一与自主Unification and Autonomy
- 生效Effective
- 2026-10-08(宣布)2026-10-08 (announced)
- 要点Key points
- Rob Leland 任首位首席人工智能官,负责确保 AI 在全实验室快速且负责任地落地;其此前负责核威慑系统中心。公共部门样本。Rob Leland becomes the first chief AI officer, responsible for fast and responsible AI adoption across the laboratory; he previously ran the nuclear deterrence systems centre. A public-sector case.
- 来源Source
- 实验室官网公告(2026-10-08)Announcement on the laboratory's official site (2026-10-08)
- 证据Evidence
- 多源已核Multi-source verified
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-015 →Spotted an error? Report GOV-015 →
GOV-016 Constellation SoftwareConstellation Software任命首席 AI 官并进入领导团队Chief AI officer appointed to the leadership team 统一与自主Unification and Autonomy2026-10-21(宣布)2026-10-21 (announced)
- 主体Entity
- Constellation SoftwareConstellation Software
- 类别Category
- 组织与职责Organization and ownership
- 对应的关Gate
- 统一与自主Unification and Autonomy
- 生效Effective
- 2026-10-21(宣布)2026-10-21 (announced)
- 要点Key points
- Daan Dijkhuizen 出任,保留 Topicus 运营集团首席执行官与风投部门管理合伙人职务;职责含以风投部门投资 AI 原生垂直软件,并支持各业务单元应用 AI 且配质量、安全与合规管控。Daan Dijkhuizen takes the role while remaining CEO of the Topicus operating group and a managing partner of the venture arm; the remit includes investing in AI-native vertical software through the venture arm and supporting business units' AI adoption with quality, safety and compliance controls.
- 来源Source
- 官方新闻稿(GlobeNewswire,2026-10-21)Official press release (GlobeNewswire, 2026-10-21)
- 证据Evidence
- 多源已核Multi-source verified
- 核查Verified
- 2026-10-23
- 状态Status
- 在用Active
- 更正记录Corrections
- 暂无更正。None so far. 发现错误?报告 GOV-016 →Spotted an error? Report GOV-016 →
编号规则、证据等级与更正政策见Numbering rules, evidence grades and the corrections policy are set out in 《记录格式说明》the Record Format note。.

