Introduction: a new safety baseline for L3 and L4 vehicles
An automated-driving demonstration can look convincing, yet reveal relatively little about how the vehicle handles unfamiliar traffic, sensor faults, a disengagement, or a situation outside the system’s intended operating conditions. For overseas buyers, fleet operators and companies evaluating Chinese vehicles or automated-driving technology partners, the central sourcing question is not whether a system can drive itself on one route. It is what it is designed to do, under which conditions, and how the manufacturer demonstrates that it can do so safely.
China approved mandatory national standard GB 44721-2026, Intelligent Connected Vehicles—Safety Requirements for Automated Driving Systems, on July 30, 2026. The official national standards database records July 1, 2027 as its implementation date. The Ministry of Industry and Information Technology (MIIT) says the standard applies to M- and N-category vehicles equipped with Level 3 (conditional automation) or Level 4 (high automation) systems and does not apply to automated parking systems.
The standard gives sourcing teams a concrete reason to revise technical questionnaires, contract specifications and supplier-verification plans before placing orders for covered products. It is also an important boundary marker: it should not be applied indiscriminately to Level 2 driver-assistance products, all passenger vehicles, or automatic parking functions.
What the standard covers—and what it does not
GB 44721-2026 is a mandatory Chinese national standard for the safety of covered automated-driving systems. It is not simply a general technical recommendation. As of October 2026, however, its July 2027 implementation date is still ahead; describing it as already effective would be inaccurate.
The MIIT announcement identifies four broad pillars: manufacturer lifecycle safety assurance, safe performance of dynamic driving tasks, human–machine interaction and user information, and a framework for conformity testing. It also identifies simulation, closed-site and on-road testing as complementary verification methods.
Level 3 means conditional driving automation within the system’s specified conditions, including a role for the human fallback driver when required. Level 4 means high driving automation within its defined operational domain, without expecting a human to perform fallback in that domain. This functional difference should be reflected in purchasing specifications, driver/operator instructions and the safety case. The manufacturer’s claimed automation level must be matched to the exact version of the product.
The Chinese classification of covered M- and N-category vehicles is also important. A buyer should identify whether a model is in scope by vehicle category and installed system—not by a marketing label such as ‘smart’, ‘intelligent’, ‘self-driving’, or ‘robotaxi ready’. A vehicle with conventional ADAS or Level 2 combined assistance is not automatically an L3/L4 product.
Why overseas buyers should care before 2027
A fleet procurement decision typically locks in vehicle hardware, sensors, software architecture, service capabilities, training and spare-part arrangements well before delivery. If an automated-driving system changes after the evidence is reviewed, the original safety demonstration may no longer fully describe the delivered configuration.
For an overseas company acquiring a Chinese vehicle for use outside China, the new Chinese standard is useful evidence of the manufacturer’s engineering and assurance approach. It does not, on its own, establish compliance with the destination market’s road-use permissions, vehicle type approval, data-protection obligations, insurance rules or operational licensing. Local regulatory approval remains a separate workstream.
The standard also differs from China’s other 2026 automotive initiatives. Broader measures on vehicle production consistency and quality oversight address manufacturing controls across road vehicles; GB 44721-2026 is specifically concerned with the safety requirements and verification of covered L3/L4 automated-driving systems.
1. Confirm the exact vehicle, system and operational design domain
Ask the manufacturer for the precise vehicle category, model identifier, automated-driving system name, software build and hardware configuration. Clarify the operational design domain (ODD): approved road types, speeds, weather, lighting, geographical restrictions, traffic situations and any other limitations claimed by the manufacturer.
Ask what the vehicle does when a proposed journey or situation falls outside that domain. The answer should distinguish preventing activation from exiting an active automated-driving mode safely. Generic statements such as ‘works in all conditions’ should be treated as marketing claims until independently supported.
Practical evidence: technical specification, version-controlled configuration list, ODD description, permitted use cases, operational restrictions and a sample user manual. Buyers should record the exact product variant each document covers.
2. Examine the manufacturer’s lifecycle safety assurance
MIIT describes GB 44721-2026 as requiring manufacturers to establish safety assurance capabilities spanning design and development, production and post-deployment activities. It specifically refers to safety policy, risk management, safety assurance and safety improvement.
Request a plain-language explanation of who owns system safety; how hazards are identified and tracked; how safety-relevant design changes are approved; and how field incidents or near misses are investigated. Ask how the manufacturer ensures that vehicle production continues to match the configuration assessed in development testing.
A supplier may provide a controlled safety-management summary rather than its full proprietary engineering file. That can be reasonable, but the buyer should still be able to understand the governance, evidence owners and scope of assurance.
3. Check dynamic-driving-task performance and minimum-risk behavior
The MIIT announcement says the standard addresses the safe execution of dynamic driving tasks and the triggering and execution of a minimum-risk strategy. This matters because a vehicle should not be evaluated only by how well it performs in normal flow.
Ask the supplier to explain how the system responds to a sensor failure, road obstruction, emergency vehicle, unexpected pedestrian, construction change, loss of positioning or other scenario relevant to its claimed ODD. Which conditions trigger a minimum-risk response, and what is that response in the context of the vehicle and road?
Do not infer specific numerical safety thresholds, minimum stopping distances or prescribed maneuvers from the MIIT announcement alone. Those details must be grounded in the applicable clauses of the complete standard and in product-specific engineering evidence.
4. Review activation, disengagement and user communication
An automated-driving system must communicate clearly when it is ready, when it is active, when it is exiting automation and what the user is expected to do. MIIT also highlights the safe activation and exit process, user-facing explanations of automation level and limitations, and Level 3 monitoring of the driver’s ability to take control.
A practical demonstration should cover the vehicle’s displayed status, alerts, language and instructions, including how a user or fleet operator learns that the system is unavailable. For Level 3 products, ask how readiness for takeover is monitored, what prompts are given and what occurs if the intended human response is not available.
Pay special attention to imported vehicles, because language adaptation and fleet training can alter the real-world effectiveness of notices even when the underlying vehicle design is unchanged.
5. Ask for a traceable and representative testing package
MIIT explains that the standard uses a combination of manufacturer assurance checks, review of a product safety file and confirmation testing. It identifies field/closed-site, road and simulation verification routes and references GB/T 41798, GB/T 44719 and GB/T 47025 as related test methods.
For sourcing purposes, request a map of the evidence rather than a single demonstration video: test configuration, scenarios, dates, versions, testing organization, environment, relevant limitations and unresolved issues. Where third-party confirmation testing applies, ask what was actually tested and by whom.
Simulation helps explore scenario coverage, but it does not replace all physical validation. Likewise, a controlled-site test does not establish that every public-road situation has been addressed. Buyers should look for the reasoning connecting the ODD, identified hazards, validation scenarios and results.
6. Verify software, sensors and production consistency
Automated-driving capability depends on integrated hardware and software. A change to cameras, radar, lidar, compute platform, actuators, calibration, sensors, maps, firmware or core decision software may change performance or invalidate assumptions underlying previous tests.
Ask for a versioned bill of materials or relevant configuration baseline, software release identifiers, change-control policy and a clear commitment to notify the buyer before safety-relevant substitutions. For fleet orders, define whether all units will arrive with the same approved system release, and how updates will be validated and documented.
An independent production check can compare vehicle identification, safety-critical hardware, labels, software identifiers where accessible, and operational behavior against approved evidence. It cannot, by itself, prove the system meets the entire standard.
7. Distinguish Chinese safety evidence from destination-market authorization
Some overseas buyers operate internationally, and a vehicle may be technically capable in one market but not authorized for the same operation in another. Before signing, map the proposed deployment country, route type, human supervision model, operational permits, insurance obligations, telecommunications and data requirements.
Ask whether the supplier has evidence relevant to the destination country and what must still be obtained. Separate compliance with GB 44721-2026 from any applicable United Nations regulations, national type-approval procedures, road-use permissions and local testing or data rules. Do not describe the Chinese standard as an international operating licence.
8. Assess what happens after delivery
For high-automation vehicles, ongoing support is part of the procurement risk. Obtain commitments for safety notices, bug fixes, incident escalation, fleet diagnostics, maintenance, training, spare parts, software-support duration and a controlled update process.
Ask who can suspend a vehicle or an automation feature after a serious safety issue is identified. Establish how the supplier will communicate material limitations and whether documentation is updated when functionality changes. For commercial fleets, plan operational contingencies if an automated-driving feature must be disabled.
A buyer’s pre-order evidence matrix
Identity: vehicle category, manufacturer, model and intended market, with no ambiguity over the contracting entity.
Automation claim: Level 3 or Level 4 classification, installed system and explicit ODD, connected to product-version identifiers.
Safety assurance: manufacturer’s policy, hazard/risk management, test governance and lifecycle improvement process.
Behavior: dynamic-task performance, activation/exit states, minimum-risk strategy and, for Level 3, takeover-readiness arrangements.
Testing: safety file, confirmation-test scope, simulation/track/road evidence, scenario coverage and known limitations.
Change control: critical hardware/software baseline, component substitutions, software-release management and post-deployment update obligations.
Market entry: Chinese compliance pathway and separate destination-market approvals, licensing, insurance and training needs.
Commercial safeguards: acceptance criteria, documentation deliverables, incident handling, service levels and support period.
Common assumptions to challenge
‘The vehicle has autonomous features, so the L3/L4 standard applies.’ Not necessarily. First establish the automation level, vehicle class and relevant system.
‘The standard was published in July 2026, so every new vehicle must already comply.’ Publication and effective dates are different; the national database gives July 1, 2027 as the implementation date.
‘A successful demonstration proves standard compliance.’ A demonstration shows performance in observed conditions, not full coverage of hazards, scenario types and assurance requirements.
‘One approval works internationally.’ China-side safety compliance does not automatically confer legal permission to operate abroad.
‘The test report covers the whole fleet.’ First verify that the tested hardware, software, ODD and manufacturer correspond to the vehicles being delivered.
Limitations and uncertainty
This article is based on the official MIIT publication and national standards catalogue, and is intended to support buyer due diligence rather than give a clause-by-clause technical interpretation. The full final standard, relevant certification/assessment procedures and product-specific test evidence should be consulted for any binding conformity determination. The article does not claim that an arbitrary supplier has passed a test, obtained an approval or achieved a specified safety performance level.
The automotive regulatory environment may continue to develop between publication and July 2027. Sourcing teams should verify the latest implementation guidance, regulatory approvals and destination-market rules at the point of contract and deployment. Some related testing standards are recommended GB/T documents; that designation should not be confused with the mandatory status of GB 44721-2026.
Conclusion
GB 44721-2026 gives buyers a clearer basis for asking what an L3 or L4 automated-driving system can actually do—and how the manufacturer supports its claims. The best procurement response is not to add one more checkbox to a quotation form, but to link the exact vehicle and software version to a defined operating domain, hazard controls, human interaction, a credible testing record and an ongoing change-management process.
For an overseas fleet or vehicle buyer, those checks should sit beside local road-use authorization, insurance, service and commercial protections. Evidence-based sourcing is particularly important when the system can take over the driving task.
Sources / Research Notes
1. Ministry of Industry and Information Technology (MIIT), ‘Mandatory national standard Intelligent Connected Vehicles—Safety Requirements for Automated Driving Systems officially released’, published August 4, 2026. Confirms approval on July 30, scope (M/N categories with L3/L4), exclusion of automated parking, proposed July 1, 2027 implementation, four safety dimensions and related verification methods. https://www.miit.gov.cn/xwfb/gxdt/sjdt/art/2026/art_16d1319a933d4ffd8501e60dc4d88491.html
2. State Administration for Market Regulation / Standardization Administration national standards database, GB 44721-2026 catalogue entry. Lists GB 44721-2026 as mandatory, published July 30, 2026, with implementation July 1, 2027; distinguishes the recommended GB/T 44721-2024. https://openstd.samr.gov.cn/bzgk/std/std_list?p.p1=0&p.p2=44721&p.p90=circulation_date&p.p91=desc
3. Related methods cited by MIIT: GB/T 41798 (site testing), GB/T 44719 (road testing) and GB/T 47025 (simulation testing). These are cited as contextual recommended test methods rather than a claim that an arbitrary vehicle has completed all tests.
Internal traceability
Research ID: SCC-RES-2026-040
Story ID: SCC-INS-2026-029
