⚠️ Editorial note: The open source ecosystem in China operates under a distinct institutional framework — characterized by state-led initiatives, intranet-like boundaries, and top-down governance. Readers should be aware that this context differs from the community-driven open source model common in other regions. The term “open source” as used in Chinese media may refer to practices that diverge from the conventional definition.
China Open Source Daily — 2026-11-07
🏛️ Policy & Regulation — The Identity Code Becomes a Condition of Market Access: MIIT Notice No. 42 and the Standard It Points At
1. The instrument, and its paper trail. On 2026-10-24 the General Office of the Ministry of Industry and Information Technology issued 《关于加强智能体终端设备进网管理的通知》, 工信厅信管〔2026〕42号 — the Notice on Strengthening Network-Access Management for Agent Terminal Equipment. It was published on MIIT’s website on 2026-11-06, picked up the same day by China News Service, People’s Posts and Telecommunications News, Sina Tech and Wallstreetcn, and reproduced in full by CAICT’s own account (中国信通院CAICT). The document defines an “agent terminal” (智能体终端) as “a telecommunications terminal device — mobile phone, tablet, wearable — with an agent preinstalled.” It cites six statutes as its basis: the Cybersecurity Law, Data Security Law, Personal Information Protection Law, Anti-Telecom-and-Online-Fraud Law, Telecommunications Regulations, and the Measures for the Administration of Network Access of Telecommunications Equipment.
2. The four work items, quoted at the strength the text supports. The notice orders four things, and the second and third matter most:
(二) Improving agent-terminal network-access management. “Mobile phones, tablets, wearable devices and other agent terminals with preinstalled agents must, in network-access testing, carry out agent adaptation verification according to the relevant standards; carry out security assessment of fraud risk and ex-ante evaluation of anti-fraud capability in accordance with law… Agents used on terminals should be equipped with a unique identity code by reference to the series of national standards on agent interconnection.” And: “The terminal manufacturing enterprise shall guarantee the consistency of the agent terminal before and after obtaining the network-access licence. Where major changes occur to the agent terminal, the manufacturing enterprise shall report to MIIT and carry out network-access testing as required.”
(三) Strengthening in-network upgrade management. “Where a terminal manufacturing enterprise carries out in-network agent upgrades by means of over-the-air (OTA) push, it shall submit an in-network upgrade plan to MIIT, specifying the upgrade time, scale, user group, business scenario, service content, business operating model, network and data security measures, and anti-fraud measures… After the in-network upgrade concludes, it shall submit an in-network upgrade report to MIIT, and only after MIIT organises expert evaluation and solicits opinions may it formally provide the service to users.”
The other two: adaptation verification — developers must publish a verified-compatible terminal/OS list on the product settings page, and submit verification status to app-distribution platforms before listing; and distribution management — distribution platforms must have developers declare by electronic signature whether they consent to agent invocation (“未经声明同意的,智能体不得强制调用调取应用”), must display the adaptation list, must not serve an agent to an incompatible device model, and must order rectification or delist agents that have not completed verification. Enforcement sits with MIIT and the provincial communications administrations; capacity-building sits with CAICT, which is to build an “agent version and terminal model adaptation registry” (智能体版本和终端型号适配库) and accelerate the identity-trust system and management platform.
3. The three gaps that are stated, not inferred. Every one of these is in the published text or in the coverage that read it closely, and all three change what the instrument does:
- “According to the relevant standards” (依据有关标准) and “by reference to” (参考) are not the same instruction. Wallstreetcn’s reading, published 2026-11-06, puts it exactly: the identity standards of the agent-interconnection series have been published, but “the specific adaptation-verification standard on which network-access testing relies, and the arrangements for its implementation, remain to be clarified” (进网检测所依据的具体适配验证标准及实施衔接安排有待明确). Three further standards are still being drafted: terminal-agent interaction with devices, with application services, and with other agents, plus a Technical Requirements for Terminal Agent Distribution Platforms.
- No effective date and no transition arrangement. Wallstreetcn states plainly that “the notice and the interpretation do not specify an effective date or transitional arrangements, and the notice does not provide for whether already-sold, already-licensed devices must retroactively complete adaptation verification.” Contrast the past two weeks: Announcement No. 48 carried an interim rule effective on publication day; this one carries no date at all.
- The review of the process is a post-hoc evaluation, and the text says so. The Chinese reporting is unusually direct on the consequence: “the iteration of agent functions thereby shifts from manufacturers’ self-directed push to a submission-and-approval process” (智能体功能迭代由此从厂商自主推送转为报批流程).
4. What the standard behind the code actually says, because it was published six months ago. The identity code is not a new invention. GB/Z 185.2—2026, Artificial Intelligence — Agent Interconnection — Part 2: Identity Code, was issued by SAMR and SAC with a publication and implementation date of 2026-05-22, and is the second of the seven-part GB/Z 185—2026 series this briefing examined on 2026-10-20. Its structure, per the standard’s own clause 5.1 and the Chinese-language reading of it: a hierarchical OID identifier per GB/T 26231, composed from left to right of the agent identity code OID, an identity-code version, an identity registration service provider, an identity registration requester, and a custom serial number, dot-separated. The registration-service-provider segment is allocated by the competent authority over the range “1–ZZZZZZ”, about 2 billion providers. Five management rules attach: a retired provider code shall not be reallocated; a change in an agent’s core function should trigger a new serial number (with the standard leaving “core function change” undefined); multiple concurrently-served versions must be separately coded; one identity code corresponds to exactly one agent; and the provider bears responsibility for the custom serial number’s uniqueness and traceability.
And then Annex B, which is the part that answers a question this series asked three weeks ago. The annex lists three routes for an international agent to obtain an identity code: (i) apply to a registration service provider recognised by the competent authority; (ii) an overseas entity may itself apply to become a registration service provider, and once approved receives a provider code allocated by the competent authority; (iii) an overseas entity may build its own agent identity management system in its own jurisdiction and, through bilateral or multilateral agreements, establish mutual trust and recognition with the core node of this system — under which its agents receive identity codes under this standard’s coding rules. That third route is a genuine reciprocity clause and it is worth recording as one, because it is the answer to whether interoperability is achievable by treaty rather than by adoption.
5. The sentence in MIIT’s own interpretation that frames everything. The ministry’s own standards interpretation of the series, published 2026-07-06, gives the motivating diagnosis without hedging: “although MCP, A2A and other protocols have already appeared internationally, a unified consensus has not yet formed, and they are also difficult to fully satisfy the needs of industrial development.” That is the ministry’s own statement of why a national standard was written rather than an international one joined. No criticism is required to read it; it is a factual claim about the state of the world as MIIT sees it, and it is the premise on which Notice No. 42’s market-access condition rests.
Institutional Analysis & Commentary:
A voluntary guidance document became a condition of market access without becoming a mandatory standard, and the mechanism is the sentence “referenced by the network-access test.”
The procedural history is worth setting out in order, because the elegance of it is the point. GB/Z is a guiding technical document — in China’s standards hierarchy, a recommendation, not a GB. On 2026-06-26, SAMR’s Zhu Meina explained at the press conference why: the series was issued as guidance “precisely because this is the agile standardisation arrangement for the industry-cultivation phase”, maximising compatibility across technical routes and reserving room for error, with the stated plan that identity-code standards would later convert to mandatory national standards. That conversion has not happened. What has happened is Notice No. 42, which makes an identity code a precondition of obtaining a network-access licence (进网许可证) — the licence without which a phone, tablet or wearable may not lawfully be sold in China — and routes the code through the GB/Z.
This is what happens when a rule’s legal form and its economic bite are separated. Formally, an agent vendor is still free to choose any protocol it likes; nothing is mandatory. Practically, the terminal that ships its agent cannot obtain the licence without the code. The obligation does not fall on the developer of the agent — it falls on the handset maker, who is the party with something to lose. That is the same structural move this briefing has traced in Chinese software procurement: regulate at the point where a licence is issued, not at the point where a choice is made. The state has not forbidden a protocol. It has made one protocol the price of a phone.
Now the third route in Annex B deserves its weight, because it is not decoration. The instinctive reading of an identity registry is “national scheme, closed by construction”. Annex B route (iii) says otherwise: a foreign jurisdiction can run its own agent identity system and federate with the core node by agreement, and its agents receive conforming codes. That is, in substance, a mutual-recognition proposal — the institutional form that works when the problem is coordination rather than control. Whether it is real is a question with an answer, and the answer is observable. The measure is not the annex’s existence but its use: whether any authority outside mainland China has signed such a bilateral or multilateral instrument, and whether the codes it assigns are honoured by the core node without an extra domestic registration step.
Against that, the honest reading of the “unified consensus has not formed” line. MIIT’s interpretation names MCP and A2A and says they lack consensus. That is a defensible engineering complaint — MCP is a tool-invocation interface and A2A is an agent-to-agent messaging layer, and neither specifies an identity authority, which is precisely the gap GB/Z 185.2 fills. A jurisdiction that needs to know which agent did what in a fraud dispute is not being unreasonable in wanting an identity root. But a standard written because international consensus was absent, then bound to a licence condition, produces a particular kind of outcome: interoperability obtained by market access rather than by adoption. In that arrangement an improvement (route iii is real mutual recognition) and a risk (the reference is to a body the competent authority designates) arrive in the same document.
The identification of the audit problem, stated as this series has stated it for a week. MIIT’s section (三) is the most consequential text of the week for anyone who actually runs an AI product. It converts OTA agent upgrades — the ordinary, continuous, unglamorous way software improves — into an expert-evaluated filing. Read generously, this is consumer protection: the notice’s own framing is that it restrains forced invocation and incompatible installation, protecting users’ right to know and to choose, and the anti-fraud provisions are concrete rather than rhetorical (fraud-risk assessment and ex-ante anti-fraud evaluation at the licence stage). Read institutionally, it means the agent cannot be updated when it needs updating without a queue. This is the direct counterpart of the point Liu Dayiheng of Qwen made at Wuzhen on 2026-11-03 and this briefing reported on 2026-11-06: an on-device agent is non-recallable — a defective capability cannot be pulled back — and the single remedy for non-recallability is the ability to push a fix. A regime that makes pushing a fix procedurally slower than shipping the defect has traded a known harm for an unknown one. Nobody in the document appears to have weighed that trade, and it is possible that nobody had to, because an identity code and an OTA review feel like two different departments.
Second-order effects, all predictable from the incentive structure rather than from anyone’s intentions. Adaptation verification is a fixed cost that falls on small agent developers more heavily than on large ones — a two-person team must now certify against handset models, publish a compatibility list, and file with distribution platforms, for every terminal OS revision. The custody is centralised and named: the adaptation registry is built at CAICT, with related testing institutions. And this is the fourth instrument in four weeks in which a Chinese regulator located the audit in a filing by the regulated party — Announcement No. 48’s public-domain declaration (2026-10-16), the NDRC’s incident-disclosure duty (2026-11-03), and now the adaptation-verification submission and the OTA report. Four filings by the same party to the same family of institutions is a coherent administrative design; it is simply not a verification design, and the difference is observable only at the point where a submission turns out to be wrong.
One perspective, not a verdict. The questions are narrow, and the websites that would answer them exist:
(i) When SAMR publishes the mandatory-conversion notice for the identity-code parts — the step Zhu Meina pre-announced in June — does the notice arrive before or after the adaptation-verification standard does? If the mandatory standard lands first, the industry spends months testing against an unpublished criterion.
(ii) Does the adaptation registry (智能体版本和终端型号适配库) publish its contents? A register that anyone can query is infrastructure. A register that only CAICT and the testing houses can query is a licensing queue with a database attached, and the difference will be visible in whether a third-party developer can check its own model’s status without an intermediary.
(iii) Is Annex B route (iii) used? One signed mutual-recognition arrangement with any non-mainland authority would convert a national scheme into a federated one. The annex is not the fact; the first signatory is.
(iv) What happens to an agent that already shipped and fails a newly published adaptation standard? The notice is silent on retroactivity and on already-licensed devices. Silence on retroactivity is not neutrality; it means the answer will be set by the first provincial communications administration that has to decide, and by whoever litigates it.
🏗️ Institutional Change — A Public Non-Profit as the Legal Form for an Open-Source Platform: Beijing Kaiyuan Institute and the Meta-Community It Operates
1. The registration, and why the legal form is the news. On 2026-10-20, approved by its supervising authority the Beijing Municipal Bureau of Economy and Information Technology (北京市经济和信息化局) and registered with the Beijing Municipal Civil Affairs Bureau (北京市民政局), Beijing Kaiyuan Industrial Software Research Institute (北京开元工业软件研究院) completed registration as a science-and-technology non-enterprise unit (科技类民办非企业单位). The announcement, published on the institute’s own account 北京开元工业软件研究院 and relayed by its platform account 元社区空间 on 2026-11-05/06, describes the step as the move “from the preparation stage to the formal-operations stage” and as “a key step in building a national-level open-source innovation platform in China’s industrial-software sector” (我国工业软件领域国家级开源创新平台建设迈出关键一步).
The backing is named at three levels: 国家部委 (a national ministry, unnamed in the text), the Beijing Municipal Bureau of Economy and Information Technology, and the Haidian District government. The institute describes its mandate as building a “shared ecosystem for common key technologies in industrial software” under the framing “AI-enabled, open-source-collaborative” (AI赋能、开源协同), with a stated goal of moving Chinese industrial software “from single-point technical breakthroughs to ecosystem leadership” and becoming “an important source of global industrial-software innovation and a standards-setter” (成为全球工业软件创新的重要策源地和标准制定者).
2. The architecture it describes, in its own terms — a resource pool, a curation register, and a room. The institute’s stated design is “common-technology resource pool, capability carrier (intelligent R&D platform + meta-community), guarantee mechanism”, with three industrial platforms (commercialisation operations, enterprise incubation, joint innovation) forming a “base support — service output — industrial landing” closed loop. The concrete, checkable pieces as of mid-October:
- The project curation register. A full-process mechanism of “solicitation — evaluation — entry into the catalogue — iteration” (征集—评估—入库—迭代). By mid-October it had collected 300-plus industrial-software and common-technology open-source projects, covering CAE, CAD and EDA as core technical directions.
- 元社区 (the “meta-community”), the online platform. Described as “the industrial-software shared-ecosystem platform” operated by the institute, serving enterprises, universities, research institutes, developers and industry users, aggregating software, algorithms, models, components, data and professional services, and moving work through professional evaluation, supply-demand matching, collaborative R&D and application validation. As of the announcement, the external online platform had entered internal testing (线上外网平台已开启内部测试) — in plain terms, not yet publicly open.
- 元空间, the physical carrier. An entire third floor — approximately 2,000 square metres — in Building 29 of the Siji Huigu · National Cybersecurity Industrial Park in Haidian, with 96 open workstations, six small meeting rooms, two large ones and one training room. More than 30 entities had applied to move in, spanning industry-academia-teaching platforms, pilot-testing platforms, industrial-software firms, universities and open-source communities.
- What has not happened yet. The institute says it will “comprehensively accelerate substantive operations”: build out its organisational structure and recruit senior technical talent; launch an industrial-software brand competition in November; put the meta-community platform formally online in December; and hold an industrial-software open-source ecosystem conference and the institute’s founding ceremony in December, at which it will “release the first batch of co-built open-source projects and scenario requirements”.
3. Why this is the same instrument as the notice above, in a different sector. The institute’s resource pool is described by its mechanism, not its contents: solicitation, evaluation, catalogue entry, iteration. It does not describe a forge, a git host, a licence, a maintainer list, or a contribution process. It describes a curating body deciding what enters a catalogue — and the catalogue’s stated purpose is “reusable, deployable industrial capability”, produced through “professional evaluation and supply-demand matching” rather than through the ordinary open-source mechanism of publishing source under a licence and letting anyone fork it.
That is not a criticism of the intent. Industrial software is the sector where open source is hardest, for structural reasons this series has touched repeatedly: CAD/CAE/EDA kernels are decades of accumulated numerical method, validated against physical test data that firms will not publish, sold to buyers who require certification and indemnity. A pure commons in this space has never been built anywhere, so any Chinese attempt to build one will involve a broker. The question is what the broker does.
Institutional Analysis & Commentary:
Three legal facts about the form, which are the actual subject of this entry.
(i) A 民办非企业单位 is a non-profit, but its registration is held by a supervising authority. Under Chinese law a minban fei qiye danwei — a “non-enterprise unit run by the people” — is a civil-law entity that may not distribute profits, that must have a competent professional supervisory body (业务主管单位) endorse its establishment, and that registers with civil affairs. The supervising body here is the Beijing Municipal Bureau of Economy and Information Technology; the announced backing includes Haidian District and an unnamed national ministry. In the vocabulary this series uses, that is a hybrid form: not a company, not a government agency, not a foundation. It has the non-distribution constraint of a non-profit and the authorisation of a state body, and it answers upward to a bureau. The absence of a profit motive is real and it is a genuine governance commitment. The absence of independence from the authorising bureau is equally real.
(ii) Compare the form to the foundation the sector already has. The obvious comparator is the OpenAtom Foundation, a non-profit foundation that holds project copyrights and runs graduation judgements — the process by which openEuler and OpenHarmony reached Graduation in November 2025. A foundation holds assets in trust for a charter and admits members who vote; a 民非 runs an operating programme under a supervising bureau. The distinction is not a hierarchy of virtue; it is a distinction in who holds the residual decision right. A foundation’s project can be forked and the fork is legitimate by the foundation’s own rules. A 民非’s catalogue decides entry by evaluation. Both are legitimate institutional designs; only one of them gives a stranger a mechanism to change what is in the catalogue without being evaluated by the curator.
(iii) The physical carrier is doing work the platform has not yet done. 2,000 square metres and 96 desks in a national cybersecurity park, with 30-plus applicant tenants, is not how an open-source project is usually described — because it is how an incubator is described. And incubators are the right instrument for a specific problem: industrial-software developers are small firms that need certification labs, customers, meeting rooms and a reason to be in the same building. The institute has built the coordination infrastructure first and the code publication second. That ordering may be correct: in sectors where the binding constraint is validated data and customers rather than source code, the room matters more than the repository. It also means that at the end of December, the thing to inspect is not the founding ceremony but the first batch of co-built projects — and specifically whether their source is published under an OSI-approved licence, in a forge that anyone can read, with a contribution process a stranger can use.
The policy line this institute is executing is written down and public, which makes accountability possible. Beijing’s 《北京市开源生态体系建设实施方案(2026—2028年)》 — issued by the same bureau, 京经信发〔2025〕56号, dated 2025-12-24 — states among its targets: “support basic software, industrial software and artificial intelligence in tackling common key technologies and promoting the open-sourcing of common components”; “support eligible municipal SOEs and key user units to build domestically developed products on open-source root communities, with scaled application”, with a “first open-source solution” (开源首方案) subsidy “at a certain percentage of the actual procurement amount”; “support Beijing’s open-source code-hosting platforms to improve project management, security scanning and security monitoring infrastructure for industrial software and other key fields, build professional communities”; and “build a non-profit open-source evaluation service… formulate open-source community health evaluation standards”. The institute is not an improvisation. It is a line item in a three-year municipal plan whose metrics are: 10 internationally influential projects (at least 5 internationally leading), 30 domestic star projects (at least 10 in AI), and 100 industry models deployed. A plan with numeric targets produces organisations with numeric targets, and a 300-project catalogue is the first number.
The honest institutional reading, offered as a reading. The form chosen — a supervised non-profit with a curation register, a physical incubator, and a founding ceremony scheduled before the platform is public — is the correct form for the problem as the bureau has defined it. That problem is not “Chinese industrial software lacks a commons”. It is “Chinese industrial software lacks validated, reusable components and the buyers who will accept them”, and the remedy for that is procurement, certification and co-location, all of which this institute does.
What it is not yet is an institution that produces the thing open source produces, which is a codebase whose direction can be changed by people who were not at the founding ceremony. That is not required by the plan, and a critic who demands it is demanding something the plan never promised. The useful question is therefore not whether the institute is real open source — it plainly is a real institution — but what the twelve months of December will show: whether the meta-community’s public launch exposes a project list with licences, or a catalogue with descriptions. The 300 projects are the strongest evidence available that work has begun, and the most important thing about them is not that there are 300. It is that they were solicited, evaluated and entered into a catalogue — three verbs that describe a register, and a register is an institution of curation, not of commoning.
Checkable questions, all answerable from documents the institute says it will publish in December.
(i) Of the first co-built projects released at the founding ceremony, how many publish source under an OSI-approved licence, and on which forge? A catalogue entry with a licence and a repository is a contribution to the commons. A catalogue entry with a description and a partner list is an announcement.
(ii) Is the meta-community platform’s resource retrieval open to unregistered users — including foreign ones? The announcement says public-service modules will open in December (资源检索、供需对接、社区运营). Whether retrieval requires registration determines whether this is infrastructure or a members’ club.
(iii) Who evaluates, and is the evaluation published? The register’s mechanism is “solicitation — evaluation — catalogue entry — iteration”. The whole institutional question is the middle two steps: by what criteria, by whom, and with what published reasoning. A published evaluation rubric is a governance asset; an unpublished one is discretion.
(iv) Does the institute acquire, hold, or merely reference the projects in its pool? A non-profit that holds copyrights and governs them is a custodian. A non-profit that catalogues projects it does not hold is a matchmaker. Both are useful; they are different institutions, and the answer determines whether a project in the pool can be forked out of it.
🔍 Commentary — Two Registers, One Design
First: the week’s two headlines are the same institutional species, and they were published eight days apart. One register records which agents exist and which devices they run on, administered by CAICT under a ministry notice with market access attached. The other records which industrial-software projects are reusable, administered by a supervised non-profit under a municipal plan with procurement subsidy attached. Neither register publishes source code. Both decide entry. And both are run by organisations whose authority comes from an authorising body rather than from the participation of the people the register is meant to serve.
That is not a coincidence and it is not a conspiracy. It is what an ecosystem looks like when the binding constraint is authorisation rather than coordination. In a system where no developer can sell into a hospital, a bank or a factory without a licence, a certification, or a procurement list, the institution that matters is the one that grants entry — and building one register per sector is the rational administrative response. The American equivalent is not a ministry; it is SBOM attestation and the insurance market, which is slower, messier, and produces the same artefacts (a register, an evaluation, a filed claim) with a different party’s signature on them.
Second: both registers would be strengthened, not weakened, by publishing their contents — and this is the one reform that costs no political capital. An identity register that lets a developer look up their own agent’s status without going through a testing house is better infrastructure. A project catalogue that publishes its evaluation rubric is a better catalogue. In both cases the state has already decided to collect the information; the marginal cost of disclosure is near zero and the marginal gain in credibility is large. The strongest available institutional argument this series can make is that Chinese regulators have built four filing regimes in four weeks and not one of them has announced a publication plan for what gets filed. That is an omission, not a policy, and omissions are the easiest things to fix.
Third, and this is where I would push back on the briefing’s own framing. There is a reading in which the two registers are evidence of a system consolidating control, and there is a reading in which they are evidence of a system industrialising. The documents support the second more than the first, and the reason is the specificity. Notice No. 42 names four work items and leaves the effective date blank — a regulator that wanted control would have dated it. GB/Z 185.2’s Annex B offers foreign jurisdictions a mutual-recognition route — a scheme built for exclusion would not have written one. The Beijing plan subsidises procurement of open-source commercial versions rather than mandating domestic development — a plan built to protect incumbents would have done the reverse. The pattern across all three is authorisation-heavy and enforcement-light, which is what a state looks like when it is trying to make a market exist rather than to make one behave. That reading has a falsifier too: if the December conference releases a catalogue with no licences, and if the adaptation registry turns out to be closed, then the first reading gains the evidence it currently lacks, and this paragraph becomes the thing to be corrected.
Fourth, the sentence worth keeping. Liu Dayiheng said at Wuzhen that governance is the variable to be taken seriously. Notice No. 42 is what governance looks like when it is taken seriously and lands on an artefact nobody designed for it: a phone with an agent in it is an artefact that must be updated after it ships, and every governance regime that reviews the update is a regime that has decided the fix can wait. The one thing the notice’s consumer-protection framing gets exactly right is that users deserve to know what is running on their device and to be able to refuse it. The thing it does not address is that the same users deserve the defect fixed by Thursday.
Sources: MIIT General Office — 关于加强智能体终端设备进网管理的通知, 工信厅信管〔2026〕42号 (issued 2026-10-24, published 2026-11-06) · CAICT — 工信部:加强智能体终端设备进网管理 (full text, 2026-11-06) · 工信微报 via Tencent News — 四问+一图,读懂《关于加强智能体终端设备进网管理的通知》 (official interpretation, 2026-11-06) · China News Service via Sina — 工信部:加强智能体终端设备进网管理 (full text, 2026-11-06) · People’s Posts and Telecommunications News via Sina Tech — 工信部发布新规,加强智能体终端设备进网管理 (2026-11-06) · Wallstreetcn — 智能体终端进网管理新规落地,OTA现网升级须经专家评估后推送 (2026-11-06): no effective date or transition arrangement; the adaptation-verification standard on which testing relies remains unclear; four further standards in drafting by CAICT, CESI and Huawei, Honor, OPPO, vivo, Xiaomi, Ant Group, Baidu, Alibaba; identity standards published 2026-05-22; agent audit and agent transaction standards in drafting · Fujian Provincial Department of Industry and Information Technology — full notice text (2026-11-06) · MIIT Department of Science and Technology via CERNET — 《人工智能 智能体互联》系列国家标准解读 (2026-07-06): “MCP, A2A and other protocols have appeared internationally but have not yet formed unified consensus”; seven parts; five conceptual domains; identity code allocated by registration service provider · SAMR national standards platform — GB/Z 185.2-2026 人工智能 智能体互联 第2部分:身份码 (issued and effective 2026-05-22; cites GB/Z 185.1-2026 and GB/T 26231) · GB/Z 185.2-2026 English-language rendering — clause 5.1 hierarchical OID structure: identity code OID, version, registration service provider, registration requester, custom serial number · Standard interpretation — GB/Z 185.2-2026 身份码 encoding rules, five management requirements, and Annex B’s three routes for international agents (including mutual recognition via bilateral/multilateral agreement) · Dazhong — 智能体互联国家标准大湾区落地,十方融海获智能体身份码注册授权 (2026-09-18): national OID registration centre at CESI; 200-plus industrial partners and 60-plus universities in the ecosystem · Securities Times — 中国信通院智能体标识与身份网络面向社会开放接入 (2026-10-09): identity network in trial operation, relying on trusted-interconnection centres in Beijing and Xiamen · Beijing News — 智能体标识与身份网络暨智能体可信互联中心(北京)正式启动试运行 (2026-09-18): CAICT vice-president Ao Li; “blockchain + DID + agent” route; W3C-conformant DID methods; first eight scenarios · Beijing Kaiyuan Industrial Software Research Institute — 官宣 | 北京开元工业软件研究院正式登记注册 (via 元社区空间, 2026-11-05/06): registered 2026-10-20 as a science-and-technology non-enterprise unit with the Beijing Civil Affairs Bureau, approved by the Beijing Municipal Bureau of Economy and Information Technology; national ministry + municipal bureau + Haidian District support; 300-plus projects across CAE/CAD/EDA; 2,000 sq m, 96 workstations, 30-plus applicant tenants; November competition, December platform launch and founding conference · Beijing Municipal Bureau of Economy and Information Technology — 《北京市开源生态体系建设实施方案(2026—2028年)》, 京经信发〔2025〕56号 (2025-12-24): targets of 10 internationally influential and 30 domestic star projects; “first open-source solution” procurement subsidy; industrial-software code-hosting and pilot-testing platforms · Beijing municipal “15th Five-Year Plan” digital-economy plan (2026-09): “加强工业软件攻关…建设工业软件开源社区” · Beijing municipal government — 2026 OpenAtom Ecosystem Conference (2026-06-25): Mulan licence family donated to OpenAtom; project donations spanning AI, embodied intelligence, electric power and industry