• The National Medical Products Administration issued a consultation draft on 13 November 2026 that would convert segmented pharmaceutical production from a case-by-case pilot into a standing procedure, with comments due by 30 November 2026.
  国家药监局于 2026 年 11 月 13 日发布征求意见稿，拟把药品分段生产从”一事一议”的试点转为常规程序，反馈截止 2026 年 11 月 30 日。

• The draft runs to six parts and twenty-three articles and rests on the revised Regulation for the Implementation of the Drug Administration Law, in force since 15 May 2026, while the biologics pilot it supersedes runs out on 31 December 2026.
  征求意见稿共六部分二十三条，上位依据是 2026 年 5 月 15 日起施行的新修订《药品管理法实施条例》；它所接替的生物制品试点将于 2026 年 12 月 31 日到期。

• Four categories of product may be split across sites, and the fourth — an originator drug already approved in China but made abroad, with some or all steps moved onshore — is the new addition.
  可分段生产的品种共四类，其中第四类为新增：已取得境内上市许可、在境外生产、部分或全部工序转移至境内分段生产的原研药品。

• At least one side of a sterile arrangement must hold three years of commercial sterile manufacturing experience in the same dosage form, and only two of the four relaxations listed in the 2025 No. 134 announcement are carried over.
  无菌药品安排中，委托方与受托方至少一方须具备同剂型三年以上商业化生产经验；2025 年第 134 号公告所列四项放宽情形中，仅两项被沿用。

• Cross-border commissioning opens in one direction only: a foreign Marketing Authorization Holder may move steps into China, while a domestic holder may not in principle commission manufacturing abroad, save for the fourth category.
  跨境委托只单向开放：境外持有人可以把工序放到境内，境内持有人原则上不得委托境外生产，第四类原研药除外。

• Antibody-Drug Conjugate intermediates that are not managed as drugs or as active pharmaceutical ingredients are excused from holding a drug manufacturing licence, but still face GMP compliance, inspection and a licence annotation.
  抗体偶联药物中不按药品或原料药管理的中间产品，可免于办理药品生产许可证，但仍须符合 GMP、接受检查并在许可证上载明。

• The obligations land squarely on the holder: two or more technical staff stationed on site for each product in production, one full-system GMP audit of every site each year, and a quality risk review chaired by the legal representative or the enterprise head.
  义务落在持有人身上：每个在产品种驻厂技术人员不少于两名，每年对全部生产场地至少开展一次全体系 GMP 审核，并由法定代表人或企业负责人主持质量风险会商。

China’s Segmented Pharmaceutical Production Rules Move From Pilot to Standing Procedure | 中国药品分段生产：从试点走向常规程序

Why this matters now

On 13 November 2026 the general office of the National Medical Products Administration published a consultation draft titled Announcement on Matters Concerning Segmented Production of Entrusted Drugs. Comments close on 30 November 2026 and go to a dedicated mailbox. The text is a draft. Nothing in it binds anyone until the final version is issued, and the final version may differ in wording, in scope, and in the transitional arrangements it attaches to products already in the pipeline.

What makes the draft worth reading now rather than in December is the calendar behind it. The segmented production pilot for biologics opened in October 2024 and closes on 31 December 2026. The revised Regulation for the Implementation of the Drug Administration Law has been in force since 15 May 2026. Between those two dates sits a gap: the Regulation provides the legal basis for splitting manufacturing across sites, but the operational route has been a pilot run on a case-by-case basis, with entry negotiated rather than published. The draft fills that gap by writing the criteria down.

Segmented production, in plain terms, means one product whose manufacturing steps are carried out at more than one site, sometimes by more than one legal entity, under a single quality assurance system that the marketing authorization holder owns. It is not outsourcing in the ordinary sense. The holder does not hand over responsibility along with the process; it keeps the responsibility and distributes the work. That distinction drives almost every clause in the draft.

For a foreign group, three features of the draft stand out. The first is the fourth eligible category, which for the first time gives an approved originator product made overseas a route to move some or all of its steps into China without re-registering as a new drug. The second is the direction of the cross-border rule: the door swings inward and, with one exception, not outward. The third is the carve-out for certain antibody-drug conjugate intermediates, which relieves a specific and expensive licensing burden without relieving the quality obligations that sit around it.

Each of those has a commercial consequence. A company that has been running a global supply chain with China as a finishing and packaging site can now contemplate the reverse: bulk or intermediate manufacture onshore, with the regulatory footprint that follows. A company that has been waiting for the pilot to become permanent can plan investment against published criteria rather than against a negotiating position. And a company in the antibody-drug conjugate space can stop treating a mature small-molecule intermediate as if it needed a drug manufacturing licence, provided it can satisfy the conditions attached.

The draft also narrows as it clarifies. New applicants must be the holder’s own development or a development the holder participated in end to end. The first category has no post-approval change route, which means a product that is already on the market and falls only within that category cannot use this announcement to restructure its manufacturing after the fact. Reading the draft as a general liberalisation would misread it. It is a procedural consolidation with a deliberately bounded scope.

From pilot to standing procedure

The pilot that the draft replaces was launched in October 2024 for biologics. Its logic was experimental: take a small number of products, work out how a split manufacturing chain could be supervised, and learn from the cases. That model has real advantages when the regulator is still deciding what the standards should be. It has an obvious disadvantage when a company needs to commit capital: entry depends on a conversation, not on a rule, and the outcome of the conversation is not published.

A standing procedure changes the nature of the decision. Instead of asking whether a product might be allowed, a company asks whether it meets published conditions. The answer becomes auditable, and a negative answer becomes something that can be addressed — by changing the product scope, by changing the site configuration, or by choosing a different route.

The draft is structured as six parts and twenty-three articles. The parts follow the life of an application rather than the structure of the statute: general provisions, eligible products, the application and approval route, the holder’s obligations, cross-border arrangements, and supplementary provisions. That ordering is itself informative. Cross-border arrangements sit in a part of their own, near the end, which reflects the fact that they are treated as a distinct policy question rather than as one more variation on a domestic application.

The legal foundation is the revised Regulation for the Implementation of the Drug Administration Law, effective 15 May 2026. The Regulation settled the principle: a holder that genuinely needs to may entrust qualified enterprises with segmented production, subject to a unified quality assurance system covering the whole process and every site. What the Regulation did not do was specify which products qualify, what an application looks like, how licensing and registration interact, or what happens when a site is outside the country. Those are the questions the announcement answers.

The timing is not accidental. A pilot that expires on 31 December 2026 and a standing procedure that has not yet been finalised in mid-November leave a narrow window. Companies with products in the pilot need continuity; companies outside it need to know whether to prepare an application. The thirty-day comment period is short for a text of this density, which suggests the regulator is working to a fixed schedule and expects comment to be targeted rather than exhaustive.

There is also a signal in what the draft does not do. It does not extend segmented production to every product that could technically be split. It does not create a general right to commission manufacturing abroad. It does not remove the requirement for a drug manufacturing licence except in one defined case. The consolidation is real, but it is a consolidation of a narrow permission, not a broadening of it.

For companies, the practical difference between a pilot and a procedure shows up in the diligence file. Under a pilot, a legal opinion on manufacturing structure could describe the regulator’s practice and little more; under a procedure, it can cite a condition and test the facts against it. That shift is small in words and large in what it lets a board approve, because a condition can be assessed before money is committed, while a practice can only be assessed after a conversation that may or may not happen.

The four eligible categories

The draft lists four categories of product that may be produced in segments. The list is exhaustive, and a product that does not fall within one of the four cannot use the procedure however sensible the manufacturing logic may be.

The first category is innovative drugs whose production process or facilities and equipment carry special requirements. The wording points at products where the manufacturing itself is the constraint: a process that needs dedicated equipment, an environment that cannot be shared, a step that does not fit inside a general-purpose plant. For these products, splitting the chain is not a cost exercise but a feasibility one.

The second category covers drugs and vaccines that departments under the State Council have identified as clinically urgently needed, and drugs and vaccines needed for responding to public health emergencies or for reserve purposes. This is the category that carries the public-interest rationale most visibly. It also carries the most external dependency, because the identification comes from outside the drug regulator.

The third category is multi-valent and multi-component vaccines. The logic here is structural: a combination vaccine may require several antigen components, each with its own manufacturing chain, and consolidating all of them at a single site is not always realistic. Segmentation lets the components be made where they are best made and combined where the combination is validated.

The fourth category is the new one, and it is the one most likely to change behaviour. It covers an originator drug that has already obtained a marketing authorization in China, is manufactured outside the country, and has some or all of its steps transferred onshore for segmented production. In substance this is an onshoring route for products that are already approved. The product does not become a new drug; it becomes a product with a different manufacturing footprint.

The legal gateway for the fourth category is the third item of Article 32 of the Regulation, which leaves room for other drugs designated by the national drug regulator. The draft uses that room. The significance is procedural as much as substantive: it shows that the reserved category is not decorative, and that the regulator intends to use it to solve specific problems rather than to keep a general discretion in reserve.

Two limits sit on top of the four categories, and both point in the same direction. A new application for marketing must concern a product that the holder developed itself or in whose development the holder participated throughout. A licensee that acquired a product and now wishes to restructure its manufacturing does not automatically qualify. And the first category has no post-approval change entry at all: only products within the second, third and fourth categories may apply to change an existing authorization into a segmented arrangement.

The combined effect is a procedure that favours companies with an unbroken development history and a live pipeline. For a group that in-licenses assets into China and then looks for manufacturing flexibility, the draft offers less than a first reading might suggest. For a group that develops in China, or that developed globally and holds the full development record, the draft offers a route that did not previously exist in published form.

Two-track procedure and the sterile-experience threshold

An application runs on two tracks at once, and confusing them is the most common way to lose time.

The licensing track runs through the provincial drug regulator where the site is located. The holder, or the entrusted manufacturer, applies for the issue or the amendment of a drug manufacturing licence covering the site and the process. The licence is the permission to manufacture; it is not the permission to market the product made there.

The registration track runs through the drug evaluation centre for products not yet approved, and through the post-approval change framework for products already on the market. For the latter, the holder assesses the change category under the Measures for the Administration of Post-Approval Changes to Drugs (Trial) and files accordingly. A change in manufacturing site is not a uniform event: the same physical move can fall into a different category depending on the product, the process and the validation data. The classification determines the depth of review, and getting it wrong is a substantive error rather than a formatting one.

The two tracks are not sequential in a strict sense. Licensing work can proceed while the registration question is being resolved, and in practice it must, because a registration application that names a site without a licence is incomplete. But the two tracks answer to different reviewers, and a company that treats a licence amendment as the whole project will find itself with permission to manufacture a product it is not cleared to place on the market.

On organisation, the draft sets a principle rather than an absolute rule: the same process step should, in principle, be entrusted to one manufacturer. The principle is stated in the application and recorded in the approved process. The practical effect is that a holder cannot quietly split a single step between two contractors to manage capacity, and cannot move a step between contractors without the change being visible in the approved documentation.

Intermediates carry their own requirement. Where a product passes between sites in an unfinished state, the intermediate must have a clear quality standard that has been approved. This is the clause that turns a logistics arrangement into a regulatory object. An intermediate without an approved standard is not a legitimate handover point; it is an unregulated material in transit.

The sterile threshold is the hardest condition in the draft. For sterile products, at least one of the two parties — the holder or the entrusted manufacturer — must have three years or more of commercial production experience in the same dosage form. The requirement is deliberately relational: it does not demand that both sides be experienced, but it does demand that the experience exist somewhere in the pair. A new entrant cannot be paired with another new entrant, however strong their respective quality systems may look on paper.

The draft then addresses relaxations. A 2025 announcement, numbered 134, set out four situations in which the experience requirement could be relaxed. The draft recognises only the second and the fourth: the case where the parties sit within the same group under a unified quality system, and the case of an entrusted manufacturer that has been materially involved in developing the product and in manufacturing clinical trial supplies. The first and third relaxations are not carried over.

The omission is significant and easy to miss. A company that has been relying on one of the two excluded relaxations during the pilot should assume that the reliance does not survive into the standing procedure. Whether the final text restores them is a question worth putting in a comment, because the draft as published does not.

Where the holder’s obligations land

The draft distributes work but not responsibility. The holder remains the party answerable for the product, and the obligations listed in the text are written to make that answerability operational rather than nominal.

The first is presence. For each product in production, the holder must station two or more technical staff at the manufacturing site. This is not an audit visit and not a periodic review; it is a continuous presence during production. For a foreign holder using a domestic contractor, it means either seconding staff or contracting for a resident technical function that the regulator will treat as the holder’s own.

The second is audit. At least once a year, the holder must conduct a full-system GMP audit covering all manufacturing sites. The phrase “all sites” is the operative one: an audit programme that covers the domestic contractor but not the overseas site, or that covers the main site but not a site performing a single step, does not meet the requirement.

The third is the cold chain. Where bulk drug substance or another intermediate travels a long distance, the holder must validate the transport, monitor temperature, and inspect the material on arrival. The three obligations are cumulative. Validation without monitoring produces a document; monitoring without incoming inspection produces a record of a failure discovered too late.

The fourth is governance. A product quality risk consultation must be chaired by the legal representative or the enterprise head. Placing the chair at that level is a deliberate choice: it removes the possibility of treating a cross-site quality question as a technical matter settled below the level at which trade-offs are made.

The fifth is data. At steps where humans and machines interact, video data must be collected. The requirement reflects a supervisory reality: in a segmented chain, the reviewer cannot rely on walking the line, so the record of the line becomes the evidence. Companies should expect the retention, accessibility and integrity of that video to be examined alongside the batch records.

None of these obligations is novel in substance — a manufacturer operating a single site faces equivalents — but their combination is demanding for a chain that has been cut into pieces. The audit obligation looks outward to every site; the presence obligation looks inward to a single line; the data obligation looks backward at what actually happened. A holder that runs them as three separate programmes usually discovers, at the worst moment, that the evidence each produces does not line up.

Supervision follows the same logic. Provincial regulators are to manage on a basis described as one file per enterprise and one approach per product, with at least one GMP compliance inspection per product each year. For cross-province arrangements, the two sides are to notify each other periodically of inspection results, sampling results, adverse reaction information and major changes, and to conduct joint investigations or assist each other where a major risk emerges.

The cross-province mechanism deserves attention from foreign groups because it is where delay usually appears. A product split between two provinces has two supervisory relationships, and the notification duty is what keeps them from diverging. A holder that maintains its own parallel record of what each province has been told will find the arrangement easier to manage than one that relies on the two regulators to reconcile.

Cross-border arrangements: a one-way opening

The fifth part, spanning Articles 17 to 20, deals with arrangements that cross the border. It is the part most relevant to foreign holders, and it is the part where the draft is most explicit about direction.

Inbound is permitted. A foreign holder may entrust some or all of its process steps to segmented production inside China, provided the product falls within one of the four categories. The condition is not waived by the cross-border character of the arrangement; it is applied to it. A product that is not in one of the four categories does not become eligible because the holder is foreign.

Outbound is restricted. A domestic holder may not, in principle, entrust a foreign manufacturer with segmented production. The single exception is the fourth category, the originator drug that was approved in China while made abroad and is now being brought onshore. In that case the arrangement is the reverse of the general prohibition, because the product’s history already involves an overseas site.

The asymmetry is a policy choice, and it is worth stating plainly rather than reading a motive into it. The draft opens a route into the country and keeps the route out of the country closed except where a specific product history justifies it. Companies planning a regional manufacturing footprint should not assume that the inbound permission implies a reciprocal outbound one.

A group weighing an inbound arrangement should also look at its contract architecture earlier than it usually does. Joint liability shifts the risk allocation between the holder and the domestic party, and the annual audit of overseas sites shifts the cost base. Both are easier to negotiate before a licence annotation exists than after, because after the annotation the domestic party knows it cannot walk away without a regulatory consequence.

On documentation, the draft offers a practical concession. A cross-border application may use production qualifications and marketing authorization materials issued by domestic or foreign regulators to substitute for application materials that do not fit the arrangement. This matters most where a foreign site holds approvals that have no Chinese equivalent document. The concession is about form, not about substance: it does not reduce the quality standard, it reduces the paperwork that would otherwise be impossible to produce.

Annotation differs by direction, and the difference is operational. Where a foreign holder uses a domestic entrusted manufacturer, the arrangement is recorded on the copy of the domestic party’s drug manufacturing licence. Where a domestic holder commissions a foreign manufacturer, the foreign enterprise’s name, the site address and the product are recorded on both the original and the copy of the licence. The distinction reflects who holds the licence: the annotation follows the licensed entity.

The most consequential clause concerns responsibility. Where a foreign holder commissions segmented production inside China, a domestic responsible person performs the foreign holder’s obligations and bears joint liability. Joint liability is not a formality. It means the domestic party is exposed to the same regulatory consequences as the holder for failures in the arrangement, and it changes the negotiation of any contract that gives effect to the arrangement.

The audit obligation travels with it. The domestic responsible person must conduct at least one full-system GMP audit each year covering the overseas sites as well. For a foreign group, this is the clause that turns a Chinese entity into a genuine quality gate for global manufacturing, not merely a local representative. Groups should consider whether the entity they nominate has the technical depth and the mandate to audit a site it does not control.

The ADC carve-out

Article 22 addresses a problem specific to antibody-drug conjugates. The manufacturing chain for these products includes small-molecule intermediates that are chemically well characterised, stable in handling, and controllable in quality, and that play a decisive role in the quality and efficacy of the finished product. Some of these intermediates are not managed as drugs and not managed as active pharmaceutical ingredients.

Under a literal reading of the licensing rules, a site making such an intermediate could be treated as needing a drug manufacturing licence. The draft resolves the question by excusing the site from holding one, on conditions.

The conditions are cumulative. The site must comply with GMP. It must be inspected by the provincial regulator where the holder is located. The site and the process must be recorded on the licence. And the site remains subject to routine supervision and to extended supervision. Where the site is in another province, the provincial regulator at the site must additionally issue an opinion consenting to the entrusted production.

Read together, the carve-out is narrower than a headline reading suggests and more useful than a cautious one. It removes a licensing requirement that was poorly matched to the material, and it replaces it with an inspection and annotation regime that keeps the site inside the supervisory perimeter. The site does not become unregulated; it becomes regulated through the holder’s licence rather than through a licence of its own.

For groups with conjugates in development, the practical question is whether a given intermediate meets the description. Maturity of process, stability of physicochemical properties, controllability of quality, and criticality to quality and efficacy are the stated criteria. A company that wants the carve-out should be able to document each of them, because the inspection will ask for exactly that.

The clause also has a supply-chain consequence. Conjugation and payload manufacturing are often spread across specialist sites, some of which are not pharmaceutical facilities in the ordinary sense. The carve-out allows those sites to remain in the chain without each acquiring a licence, which reduces a structural barrier to locating conjugate work in China. It does not reduce the holder’s obligation to audit, to annotate, and to answer for the result.

Vaccines and the sequencing question

Vaccines appear in the second and third categories, and the draft adds a sequencing rule for a subset of them. Where the circumstances in Article 33 of the Regulation apply, the applicant must first seek approval from the national drug regulator under the Vaccine Administration Law, and only then proceed to licensing and registration.

The sequencing matters because it inverts the ordinary order of work. In a standard application, licensing and registration run in parallel with review. For the vaccines caught by Article 33, the approval comes first, and the licensing and registration steps follow it. A project plan that assumes parallel tracks will be wrong for these products, and the error will surface late.

The draft does not restate the substance of Article 33, and companies should read the Regulation directly rather than infer the scope from the announcement. What the announcement contributes is the procedural consequence: approval under the Vaccine Administration Law is a precondition, not a parallel filing.

What foreign holders should do before 30 November

The comment window closes on 30 November 2026, and the mailbox is the one named in the notice. A comment that identifies a specific article and a specific operational consequence is more likely to be useful than a general statement of support or concern. Four questions are worth answering internally before that date.

First, does the portfolio map onto the four categories, and onto which one? A product may appear to fit the first category because its process is unusual, while in fact fitting the fourth because its history is an overseas approval followed by an onshore transfer. The categories carry different consequences, including whether a post-approval change route exists at all.

Second, is the development history clean? For a new application, the holder must have developed the product itself or participated throughout. Groups that acquire assets should establish, before planning a segmented arrangement, whether the development record supports the claim. Where it does not, the route is closed regardless of the manufacturing logic.

Third, does the sterile experience sit on the right side? At least one party must hold three years of commercial experience in the same dosage form, and only two of the four relaxations are recognised. A group relying on an excluded relaxation should test the assumption now rather than at filing.

Fourth, who will be the domestic responsible person, and can that entity audit? Joint liability plus an annual full-system audit of overseas sites is a demanding combination. The entity needs technical capability, access to the sites, and a mandate that survives commercial pressure.

Beyond the four questions, three preparatory steps are worth taking. Map every site in the proposed chain against the licence annotation rules, because the annotation differs by direction and an omission is a defect. Confirm that each intermediate handover point has an approved quality standard, since an unapproved intermediate is not a valid boundary. And for conjugates, assemble the evidence that the intermediate meets each of the four criteria in Article 22, because the carve-out is granted on documentation and inspection rather than on description.

Outlook

The draft is a procedural document, and its effects will be felt in individual applications rather than in a single announcement. The parts most likely to be adjusted before the final text are the two excluded relaxations and the treatment of products whose development history is shared between an originator and a Chinese licensee. Both are the kind of question that a comment period exists to surface.

For planning purposes, the safe assumption is that the four categories, the two-track procedure, the one-way cross-border rule and the antibody-drug conjugate carve-out will survive in substance. The variables are the transitional arrangements for products currently in the pilot, and the precise wording of the experience requirement.

Companies with pilot products should treat the 31 December 2026 expiry as a real deadline and plan continuity rather than assume an automatic rollover. Companies without pilot products now have, for the first time, a published description of what a segmented arrangement looks like — which is itself the main change. The permission was always there in principle. What was missing was the procedure, and the procedure is now on the table for comment.

Sources

• 国家药监局综合司·《关于药品委托分段生产有关事项的公告（征求意见稿）》征求意见页面（反馈截止 2026-11-30）：https://www.nmpa.gov.cn/xxgk/zhqyj/zhqyjyp/20261113111413106.html
• 中国政府网·国务院公报 2026 年第 4 号《中华人民共和国国务院令（第 828 号）中华人民共和国药品管理法实施条例》（2026-01-16 公布，2026-05-15 施行，含第三十二条分段生产条款）：https://www.gov.cn/gongbao/2026/issue_12546/202602/content_7057461.html
• 蒙城县人民政府（转载）·《中华人民共和国药品管理法实施条例》（2026 年修订，含第三十二条、第三十三条分段生产条款）：https://www.mengcheng.gov.cn/XxgkContent/show/3008349.html

---

中国药品分段生产：从试点走向常规程序

为什么此刻值得看

2026 年 11 月 13 日，国家药监局综合司发布《关于药品委托分段生产有关事项的公告（征求意见稿）》。反馈截止 2026 年 11 月 30 日，意见投递到公告指定的邮箱。这份文本是征求意见稿，在正式稿印发之前不对任何人产生约束力，正式稿的措辞、范围以及对在审品种的过渡安排，都可能与现在不同。

让这份稿子值得现在读、而不是等到十二月再读的，是它背后的时间表。生物制品分段生产试点 2024 年 10 月启动，2026 年 12 月 31 日到期。新修订的《药品管理法实施条例》2026 年 5 月 15 日起施行。两个日期之间留着一道缝：条例给了分段生产的法律依据，但落地路径一直是试点，逐案商议，入口靠谈、不靠公布。这份公告要做的，就是把标准写下来。

用平实的话说，分段生产指的是同一个产品的不同生产工序分散在多个场地完成，有时还分散在多个法人主体手里，而这些场地与主体共同处在一套由上市许可持有人拥有的质量保证体系之下。它不是通常意义上的外包。持有人并没有随工序一起把责任交出去，它保留责任，只把工作分出去。这条区分，几乎决定了公告里的每一条。

对外资集团而言，这份稿子有三处值得注意。第一是第四类品种，它第一次为”已在境内获批、在境外生产”的原研药开出一条把部分或全部工序搬到境内、又不必按新药重新注册的路。第二是跨境条款的方向：门朝里开，除一种情形之外不朝外开。第三是抗体偶联药物某些中间产品的豁免，它卸掉了一项与物料性质并不匹配、代价却很高的许可负担，但并没有卸掉围绕着它的质量义务。

这三点各有商业含义。一家长期把中国当作成品加工与包装基地的全球企业，现在可以设想反过来做：把原液或中间产品放在境内生产，并接受随之而来的监管足迹。一家一直在等试点转正的企业，可以按公布的条件而不是按谈判地位来安排投资。而一家做抗体偶联药物的企业，可以不再把某个工艺成熟的小分子中间体当作需要药品生产许可的对象——前提是它能满足附在后面的条件。

这份稿子在澄清的同时也在收窄。新申请上市的品种，必须是持有人自主研发或者全程参与研发的。第一类没有上市后变更入口，也就是说，一个已经上市、且只落在第一类的产品，不能借这份公告在事后重组生产布局。把稿子读成一次普遍放开，是误读。它是一次程序上的收口，范围被有意画得很清楚。

从试点到常规程序

这份稿子接替的试点，2024 年 10 月针对生物制品启动。它的逻辑是试验性的：挑少量品种，摸索一条被切开的制造链该怎么监管，再从个案里学。在监管方还没决定标准应该是什么的时候，这种模式有实实在在的好处。但当企业需要押上资本时，它的短板也很明显：能不能进，取决于一次谈话；谈出什么结果，不会公布。

常规程序改变的是决策的性质。企业不再问”这个产品有没有可能被允许”，而是问”它是否满足已经公布的条件”。答案变得可核查，否定性的答案也变成可以处理的东西——调整品种范围、调整场地配置，或者换一条路走。

稿子的结构是六部分、二十三条。各部分的划分顺着一个申请的流程走，而不是顺着法条的体例走：总则、可分段品种、申请与审批、持有人义务、跨境安排、附则。这个次序本身带着信息。跨境安排单独成篇、位置靠后，说明它被当作一个独立的政策问题处理，而不是境内申请的一个变体。

法律基础是 2026 年 5 月 15 日施行的新修订《药品管理法实施条例》。条例把原则定了下来：确有需要的持有人可以委托符合条件的企业分段生产，前提是建立覆盖全过程、全部生产场地的统一质量保证体系。条例没有做的是：说明哪些品种够格、一份申请长什么样、许可与注册怎么衔接、场地在境外时怎么办。这些正是公告要回答的问题。

时间上的安排并非偶然。一个 2026 年 12 月 31 日到期的试点，加上一份十一月中旬还没定稿的常规程序，中间的窗口很窄。已在试点里的品种需要连续性，试点之外的品种需要知道要不要准备申报。对一份密度这么高的文本给出三十天反馈期，说明监管方按固定日程推进，并且预期意见是点对点的，而不是面面俱到的。

稿子没做什么，也是一种信号。它没有把分段生产扩展到所有技术上可以拆分的品种；它没有创设一项委托境外生产的普遍权利；除了一种明确情形，它没有免除药品生产许可。收口是真的，但收的是一项范围很窄的许可的口，而不是把它放宽。

对企业而言，试点与程序之间的实际差别，会出现在尽职调查文件里。试点之下，一份关于生产结构的法律意见只能描述监管惯例，再多也就到此为止；程序之下，它可以引用一项条件，并拿事实去比对。这个转变在字面上很小，在”董事会能批什么”上却很大——条件可以在掏钱之前评估，惯例只能在一场不知会不会发生的谈话之后评估。

四类可分段品种

稿子列出四类可以分段生产的品种。这份清单是穷尽式的，一个品种只要不落在这四类里，无论制造逻辑多么合理，都用不上这套程序。

第一类是生产工艺、设施设备有特殊要求的创新药。这个表述指向的是制造本身构成约束的产品：需要专用设备的工艺、不能共用的环境、塞不进通用厂房的一道工序。对这类产品，拆链不是成本问题，而是可行性问题。

第二类是国务院有关部门提出的临床急需药品及疫苗，以及应对突发公共卫生事件急需或者储备需要的药品及疫苗。这是公共利益理由最外显的一类。它同时也最依赖外部：认定来自药品监管系统之外。

第三类是多联多价疫苗。这里的逻辑是结构性的。一支联合疫苗可能要由几种抗原组分构成，每种组分各有自己的制造链，把它们全部塞进一个场地并不总是现实。分段让各组分在最适合的地方生产，再在组合工艺经过验证的地方完成组合。

第四类是新加的，也是最可能改变行为的一类。它指已取得境内上市许可、在境外生产、把部分或者全部工序转移至境内分段生产的原研药品。实质上，这是一条针对已获批产品的境内落地通道。产品不会因此变成新药，它只是换了一套生产足迹。

第四类的法律入口，是《实施条例》第三十二条第三项——为”国家局规定的其他药品”留的那道口子。稿子用掉了这道口子。意义一半在程序、一半在实体：它说明这项保留条款不是摆设，监管方打算用它解决具体问题，而不是留着一项笼统的裁量。

四类之上还压着两条限制，方向一致。新申请上市的品种，必须是持有人自主研发、或者全程参与研发的。一家把品种买进来、现在想重组生产的持证企业，并不自动够格。第一类则完全没有上市后变更入口：只有落在第二、第三、第四类的产品，才可以把已有的批件改成一种分段安排。

两条限制合起来，形成一套偏向”研发史完整、在研管线在手”的企业的程序。对一个把资产引进中国、再去找制造灵活度的集团来说，这份稿子给的东西比初读时感觉的要少。而对一个在中国做研发、或者在全球做研发并握有完整研发记录的集团来说，这份稿子给的是一条此前没有以公布形式存在的路。

两条线并行与无菌门槛

一份申请同时跑两条线，把两条线混为一谈，是最常见的误时方式。

许可线走场地所在地的省级药品监管部门。持有人或者受托生产企业，就场地与工艺申请《药品生产许可证》的核发或者变更。这张证是生产的许可，不是把那里生产出来的产品拿去上市的许可。

注册线走两条分支：尚未上市的品种报药品审评中心；已经上市的品种，按《药品上市后变更管理办法（试行）》走上市后变更。后一条分支里，持有人要先判定变更类别，再据以申报。生产场地的变化不是一件整齐划一的事：同样一次物理搬迁，因为品种、工艺、验证数据不同，可能落在不同类别。类别决定审评的深浅，判错类别是实体错误，不是格式错误。

两条线在严格意义上并不是先后关系。许可工作可以在注册问题解决的同时推进，实践中也必须如此，因为一份写明了场地却没有相应许可的注册申请是不完整的。但两条线面对的是不同的审评人，把许可证变更当成整个项目的企业，最后会拿到一张”可以生产、但不能上市”的许可。

在组织方式上，稿子给的是原则而不是铁律：同一道工序原则上委托一家生产企业。这条原则要在申请中说明，并在核准的工艺中载明。实际效果是，持有人不能为了调节产能把一个工序悄悄分给两家受托方，也不能在没有留下文件痕迹的情况下把一道工序从一家换到另一家。

中间产品有自己的要求。产品以未完成状态在场地之间流转的，中间产品必须有明确、且经过核准的质量标准。这一条把一项物流安排变成了监管对象。没有核准标准的中间产品，不是一个合法的交接点，而是一批处在监管真空里的物料。

无菌门槛是稿子里最硬的条件。对无菌药品，委托方与受托方之中至少一方，必须具备同剂型三年以上的商业化生产经验。这个要求是有意设计成”关系型”的：它不要求双方都有经验，但要求这份经验确实存在于这一对组合之中。两个新手凑在一起，无论各自的质量体系在纸面上多漂亮，都不行。

稿子接着处理放宽。2025 年的一份公告（第 134 号）列了四种可以放宽经验要求的情形。稿子只认其中第二项和第四项：双方同处一个集团、实行统一质量体系的；以及主要参与该药品研发、并生产过临床试验用药品的受托方。第一项与第三项没有沿用。

这个省略很重要，也很容易被忽略。在试点期间依赖那两项被排除的放宽情形的企业，应当假定这份依赖不会带进常规程序。正式稿会不会恢复，是一个值得写进意见的问题，因为按现在公布的文本，答案是否定的。

义务落在持有人身上

稿子把工作分了出去，但没有把责任分出去。持有人始终是对产品负责的一方，文本列出的义务，是为了让这份责任可执行，而不是停留在名义上。

第一项是在场。每个在产品种，持有人要在生产场地派驻两名以上技术人员。这不是一次审计访问，也不是一次阶段性复核，而是生产期间的持续在场。对使用境内受托方的境外持有人来说，这意味着要么派人，要么安排一项常驻的技术职能，而监管方会把这项职能视为持有人自己的。

第二项是审核。持有人每年至少开展一次覆盖全部生产场地的全体系 GMP 审核。”全部场地”是这句话的关键。一套只覆盖境内受托方、不覆盖境外场地的审核方案，或者只覆盖主场地、不覆盖承担单道工序场地的方案，都不满足要求。

第三项是冷链。原液或者其他中间产品长距离运输的，持有人要验证运输、监控温度，并在入厂时检验。三项义务是叠加的。做了验证不做监控，得到一份文件；做了监控不做入厂检验，得到一份”发现得太晚”的记录。

第四项是治理。产品质量风险会商必须由法定代表人或者企业负责人主持。把主持人定在这一层是有意的选择：它排除了把跨场地质量问题当成技术细节、在做出取舍的那一层之下解决的可能。

第五项是数据。在人机交互的环节，要采集视频数据。这条要求反映的是一种监管现实：在一条被切开的链上，检查员没法靠走线来确认，于是产线的记录就成了证据。企业应当预期，这些视频的留存、可调取性与完整性，会和批记录一起被看。

这几项义务单拎出来都不新鲜——只经营一个场地的生产企业面对的是同类要求——但叠在一条被切开的链上，要求就变得苛刻。审核义务朝外看，看的是每一个场地；在场义务朝内看，看的是某一条产线；数据义务朝后看，看的是当时到底发生了什么。把三者当成三个互不相干的项目的持有人，往往会在最糟糕的时候发现：三套证据对不上。

监督遵循同样的逻辑。省级监管部门按”一企一档、一品一策”管理，每个品种每年不少于一次 GMP 符合性检查。跨省安排的，两地要定期互相通报检查结果、抽检结果、不良反应信息与重大变更；出现重大风险的，联合调查或者相互协查。

跨省机制值得外资集团多看一眼，因为拖延通常就出在这里。一个跨两省的产品有两段监管关系，而通报义务是防止两段判断分叉的东西。自己另建一份”哪个省被告知了什么”的台账的持有人，会比指望两地监管方自行对账的持有人更省事。

跨境安排：只朝一个方向开

第五部分（第十七条至第二十条）处理跨越关境的安排。这是与境外持有人最相关的一部分，也是稿子把方向讲得最明白的一部分。

进来可以。境外持有人可以把部分或者全部工序委托境内分段生产，前提是品种落在四类之中。这个条件不会因为安排具有跨境性质而免除，而是被适用到跨境安排上。不在四类里的品种，不会因为持有人是境外的就变得够格。

出去受限。境内持有人原则上不得委托境外生产企业分段生产。唯一的例外是第四类——那个在境内获批、在境外生产、如今正在往境内搬的原研药。这种情形下，安排本身就是一般禁令的反面，因为产品的历史里本来就有一个境外场地。

这种不对称是一项政策选择，值得平实地陈述，而不必去揣测动机。稿子开了一条进国的路，同时把出国的路关着，只在特定产品历史足以支撑时例外。规划区域制造布局的企业，不应当从”进来可以”推出”出去也可以”。

考虑把工序搬进来的集团，还应当比惯常更早去看合同架构。连带责任改变的是持有人与境内一方之间的风险分配，对境外场地的年度审核改变的是成本结构。两件事在许可证载明之前谈，都比在载明之后谈容易——载明之后，境内一方清楚自己没有退路，因为退出本身就是一项监管后果。

在材料上，稿子给了一项务实的让步。跨境申请可以用境内、境外监管部门出具的生产资质与上市许可材料，替代那些不适用于该安排的申请材料。这在境外场地持有境外批件、而该批件在境内没有对应文件时最有用。这项让步是形式上的，不是实体上的：它不降低质量标准，只是减少了本来无法产出的那部分纸面工作。

载明方式按方向不同，而这个差别是操作性的。境外持有人委托境内生产的，在境内受托方《药品生产许可证》副本上载明。境内持有人委托境外生产的，境外企业名称、场地地址、产品要在许可证正本与副本上同时载明。区别的来源是谁持证：载明跟着持证主体走。

最有分量的条款是责任。境外持有人跨境委托境内分段生产的，由境内责任人履行境外持有人的义务，并承担连带责任。连带责任不是手续。它意味着境内一方要为这项安排中的失误承担与持有人同等的监管后果，也意味着任何落实这项安排的合同，谈判方式都要改。

审核义务随之同行。境内责任人每年至少要做一次覆盖境外场地的全体系 GMP 审核。对境外集团来说，这一条把一家中国主体变成了全球制造体系里一道真实的质量关口，而不只是一个本地代表处。集团需要考虑：拟指定的这家主体，有没有技术深度和授权去审核一个它并不控制的场地。

抗体偶联药物的例外

第二十二条处理的是抗体偶联药物特有的一个问题。这类产品的制造链里有小分子中间体，化学性质清楚、处置中稳定、质量可控，并且对成品的质量与疗效起关键作用。其中一些中间体既不按药品管理，也不按原料药管理。

按许可规则的直白读法，生产这种中间体的场地可能被当作需要药品生产许可证。稿子用”附条件豁免”回答了这个问题。

条件是叠加的。场地要符合 GMP；要经持有人所在地省级监管部门检查；场地与工艺要在许可证上载明；场地仍要接受日常监管与延伸监管。场地在另一省的，还须由场地所在地省级监管部门出具同意受托生产的意见。

合起来读，这项例外比标题式读法要窄，也比保守式读法要有用。它卸掉了一项与物料性质并不匹配的许可要求，换上一套检查加载明的安排，把场地留在监管边界之内。场地不会因此变成无人监管，它变成通过持有人的许可证接受监管，而不是自己持一张证。

对有偶联药物在研的集团来说，实际问题是一项给定的中间体是否落入这个描述。工艺成熟、理化性质稳定、质量可控、对质量疗效起关键作用，是文本给出的四项标准。想要用上这项例外的企业，应当能为每一条留下文件，因为检查要的正是这些。

这一条还有供应链上的后果。偶联与载荷的生产常常分散在专业场地之间，其中一些并不是通常意义上的制药车间。例外让这些场地可以留在链上而不必各自去拿一张证，从而降低了把偶联环节放在境内的结构性障碍。它不降低持有人的审核、载明与最终担责义务。

疫苗与先后顺序

疫苗出现在第二类和第三类里，稿子又为其中一个子集加了顺序规则。符合《实施条例》第三十三条情形的，要先按《疫苗管理法》向国家药监局申请批准，再办许可与注册。

这个顺序之所以重要，是因为它把通常的工作次序倒了过来。在一般申请里，许可与注册同审评并行推进。对落入第三十三条的疫苗，批准在前，许可与注册在后。按并行双线做的项目计划，在这些品种上是错的，而错误会在很晚的时候才暴露。

稿子没有复述第三十三条的实体内容。企业应当直接去读条例，而不是从公告反推适用范围。公告贡献的是程序后果：按《疫苗管理法》取得批准，是前置条件，不是一条并行的申报。

境外持有人应在 11 月 30 日之前做什么

反馈窗口 2026 年 11 月 30 日关闭，邮箱是公告里写明的那个。点名到具体条文、并说明具体操作后果的意见，比泛泛的支持或担忧更有用。在此之前，有四个问题值得先在内部回答清楚。

第一，产品组合落在四类的哪一类？一个产品可能看起来像第一类，因为工艺特殊；实际上却属于第四类，因为它的历史是”境外获批、再往境内搬”。不同类别的后果不同，包括到底有没有上市后变更入口。

第二，研发史是否干净？新申请上市，持有人必须自己研发、或者全程参与研发。收购资产的企业，在规划分段安排之前，应先确认研发记录能否支撑这个主张。支撑不了，无论制造逻辑多顺，路都是关着的。

第三，无菌经验是否落在对的一侧？至少一方要有同剂型三年以上商业化生产经验，而四项放宽只有两项被承认。依赖被排除的那两项的企业，现在就应当检验这个假设，而不是等到申报。

第四，境内责任人是谁，它审得动吗？连带责任加上对境外场地的年度全体系审核，是很硬的一组要求。这家主体需要技术能力、需要能进入那些场地、还需要一份不会被商业压力磨掉的授权。

四个问题之外，还有三件准备工作值得做。把拟设链条上的每个场地对照许可证载明规则过一遍，因为载明方式按方向不同，漏一处就是缺陷。确认每一个中间产品交接点都有核准的质量标准，因为没有核准标准的中间产品不是一个有效的边界。做偶联药物的，把中间体满足第二十二条四项标准的证据攒起来，因为这项例外是凭文件与检查给出的，不是凭描述给出的。

展望

这是一份程序性文件，它的效果会体现在一个个具体申请里，而不是体现在某一次发布里。正式稿最可能调整的地方，是那两项被排除的放宽情形，以及研发史由原研方与境内被许可方共享的品种怎么处理。这两类问题，正是反馈期存在的理由。

做规划时，稳妥的假设是：四类品种、双线程序、单向跨境、抗体偶联药物例外，在实体上会保留下来。变量在于试点内品种的过渡安排，以及经验要求的具体措辞。

手上有试点品种的企业，应当把 2026 年 12 月 31 日当作真实期限，按”需要接续”来准备，而不是假定会自动顺延。没有试点品种的企业，第一次拿到了一份关于”分段安排长什么样”的公开描述——这本身就是主要的变化。许可在原则上一直存在。缺的是程序，而程序现在摆上了桌面，等着被评论。

来源

• 国家药监局综合司·《关于药品委托分段生产有关事项的公告（征求意见稿）》征求意见页面（反馈截止 2026-11-30）：https://www.nmpa.gov.cn/xxgk/zhqyj/zhqyjyp/20261113111413106.html
• 中国政府网·国务院公报 2026 年第 4 号《中华人民共和国国务院令（第 828 号）中华人民共和国药品管理法实施条例》（2026-01-16 公布，2026-05-15 施行，含第三十二条分段生产条款）：https://www.gov.cn/gongbao/2026/issue_12546/202602/content_7057461.html
• 蒙城县人民政府（转载）·《中华人民共和国药品管理法实施条例》（2026 年修订，含第三十二条、第三十三条分段生产条款）：https://www.mengcheng.gov.cn/XxgkContent/show/3008349.html