Advanced driver-assistance marketing has long focused on experience: lane changes, highway ramps and fewer takeovers. After August 4, automakers must answer a harder question: can those functions pass a national mandatory standard covering safety files, simulation, closed-course tests and road tests?
China’s Ministry of Industry and Information Technology organized the new standard, GB 44721-2026, Safety requirements for automated driving systems of intelligent connected vehicles. It is scheduled to take effect on July 1, 2027. ITHome and Yicai both verified the release. It is China’s first mandatory national standard for L3 conditional automation and L4 high automation, upgrading the 2024 recommended standard GB/T 44721-2024 into a binding access framework.
Four Lines Define The Boundary
The standard applies to M-category passenger vehicles and N-category goods vehicles equipped with L3 or L4 automated driving systems. Automated parking systems are outside its scope. That boundary matters because marketing phrases such as urban NOA or highway navigation must now be mapped back to formal automation levels and vehicle classes.
Public materials describe four requirement lines: lifecycle safety management by manufacturers, dynamic driving task execution, human-machine interaction and user disclosure, and multidimensional testing. In plain terms, an automaker must prove not only that a car can drive itself in a scenario, but also that development, production, post-deployment improvement, user prompts and third-party checks form a closed loop.
Takeover And Exit Become Hard Requirements
The riskiest L3/L4 moments often appear at the edge of automation: the system exits, the driver must take over, or the vehicle must enter a minimum-risk state. The new standard puts these moments into mandatory requirements.
Official explanations say the system’s safety level should at least match a qualified and attentive driver performing the dynamic driving task, while avoiding unreasonable risks to users and other road users. It also requires clear status signals for readiness, activation and exit, and driver takeover-capability monitoring for L3 systems.
Safety Files Change The Cost Structure
Yicai reported one telling detail: the standard includes a 17-page safety-file appendix. A safety file is not a consumer specification sheet. It is closer to a responsibility chain submitted to regulators and testing bodies: function boundaries, risk identification, test coverage, monitoring and post-deployment improvement.
That changes the economics of high-level smart driving. Strong algorithms and road-test fleets still matter, but safety cases, scenario libraries, simulation resources, third-party testing and continuous updates become production costs. Full-stack automakers may find it easier to collect evidence and close the loop, while brands relying on external suppliers must clarify supplier responsibility, vehicle responsibility and user disclosure.
A More Auditable Phase
For the AI industry, this is a regulatory-method story. Automated driving is one of the earliest AI systems to face real-world public risk. Model decisions, sensor fusion, planning, control and human-machine interaction all affect road safety. China’s standard does not look only at model capability; it combines enterprise process, system behavior, user information and testing methods.
The next proof points are concrete: how automakers update L3/L4 user manuals, how third-party testing bodies run confirmatory tests, whether local pilots adjust access rules, and whether cars already marketed as advanced smart-driving products change their usable scope or takeover prompts.
Sources: Ministry of Industry and Information Technology, National Public Service Platform for Standards Information, ITHome, CocoLoop, Yicai; sources verify GB 44721-2026 status, implementation node, L3/L4 scope, M/N vehicle boundary, four safety requirement groups, related confirmatory-test standards and safety-file appendix.