A Guide to Interpreting China's L3/L4 Automated Driving Regulation

A Guide to Interpreting China's L3/L4 Automated Driving Regulation

On July 30, China's first mandatory national standard for L3 (conditionally automated driving) and L4 (highly automated driving) systems—Intelligent and Connected Vehicles—Safety Requirements for Automated Driving Systems (GB 44721—2026)—was officially released. This new national standard will take effect on July 1, 2027, establishing a unified "safety access threshold" for autonomous vehicles. One of the biggest changes in this standard is that it has been upgraded from the recommended national standard (GB/T 44721—2024) issued in 2024 to a mandatory one—shifting from "recommended practice" to "required compliance," with legally binding regulatory force. After July 1, 2027, all new vehicle applications for public announcement and new vehicle launches must meet the standard; otherwise, sales will be prohibited.

The new standard primarily governs M-class (passenger) and N-class (freight) vehicles equipped with L3 and L4 automated driving systems. L0 to L2 "driver assistance" systems, which still require human supervision at all times, are not covered. In addition, automatic parking systems are also exempt from this standard.

In simple terms: if your car only has features like adaptive cruise control or lane-keeping assist, this standard has nothing to do with you. But if you buy a car that can drive itself under specific conditions (L3), or can fully drive itself in a specific area (L4), then that vehicle must comply with the new standard's requirements.

In the past, when an accident occurred, investigators primarily looked at driver operation. From now on, things are different—if an automated driving system is involved in an accident, regulators, insurers, and courts will first scrutinize the manufacturer's algorithms, data, and safety management processes. The new standard requires manufacturers to establish safety assurance capabilities across the entire lifecycle—from design and development, through manufacturing, to post-deployment—across four dimensions: safety policy, risk management, safety assurance, and safety improvement. These four areas are defined as follows—Safety policy: objectives set to achieve the intended safety outcomes; Risk management: management processes used to identify, assess, and mitigate risks at the organizational, personnel, and technical levels; Safety assurance: periodic independent internal audits and external inspections as evidence; Safety improvement: a continuous improvement process.

The new standard imposes hard requirements on the automated driving system itself: the safety level of the automated driving system must at least match that of a competent and attentive human driver. When the system encounters situations it cannot handle, it must be able to safely disengage without allowing risks to persist. Specifically: when the system experiences failures such as sensor faults, positioning faults, or computing platform hardware/software failures, it shall actively enter a minimal risk state, promptly issue a takeover request via audible/visual or combined means, and actively apply braking control, decelerate to a stop, or pull the vehicle over to a safe location. In plain terms: if the system cannot handle the situation, it cannot simply give up—it must bring the vehicle to a safe stop by itself. In addition, L3 automated driving systems must also be equipped with a driver takeover capability monitoring function—that is, when the system needs the human to take over, it must first confirm that the driver is actually ready. Here is a "life-saving rule" that every car owner should know: the standard stipulates that after issuing a takeover request, the system must ensure the driver has at least 10 seconds of response time. Within 4 seconds after the request is issued, the system must escalate the alert—adding audible alarms and steering wheel/seat tactile vibrations on top of the dashboard visual prompt. More importantly, when it comes to liability: if you fail to take over within 10 seconds after the system issues the request, liability shifts from the manufacturer to you. Whether the system adequately alerted you will also be a factor in determining whether you should be held responsible.

What the new standard aims to solve is transforming case-by-case approvals into a standardized pathway that all manufacturers can follow.

1.Once manufacturers build robust safety systems, conduct sufficient testing, and pass inspections, they can apply for market access. For manufacturers truly investing in L3 and L4, uncertainty is significantly reduced. The new standard establishes a three-part inspection framework: "enterprise capability verification + safety case verification + confirmatory testing." What is a safety case? The new standard devotes a dedicated 17-page appendix (Appendix D) to specifying the requirements for compiling the safety case.

2.Before launching a new vehicle, manufacturers must submit a complete set of structured safety materials to regulators, including: functional description (ADS configuration and operational characteristics, activation/deactivation conditions, etc.), system architecture and principles (hardware/software component overview, relationships between ADS and other vehicle systems), identified ADS-related risks and how they are mitigated/reduced/accepted, hazard analysis and risk assessment summary, and verification criteria and validation targets.

3.This framework draws on approaches from aviation, rail transit, and nuclear power—requiring manufacturers to present a three-tier chain of "claim—argument—evidence," submitting complete data from simulation, proving ground, and real-road testing to demonstrate that system risks are under control. Moreover, the safety case is not a one-time submission—OTA updates must also be accompanied by updated safety case documents: first update, then update the safety case, then deploy on the road. In the confirmatory testing phase, third-party testing institutions will employ three comprehensive methods—proving ground testing (GB/T 41798), road testing (GB/T 44719), and simulation testing (GB/T 47025)—to conduct a full-spectrum evaluation of the automated driving system.

The new standard will officially take effect on July 1, 2027. The one-year period before that serves as a transition period, during which manufacturers must improve their safety systems and technical capabilities to meet the new requirements. It is understood that major manufacturers including Huawei's automotive partners, XPeng, GAC, Zeekr, and BYD have all set 2025 to 2027 as the critical window for mass production of L3 and L4 vehicles. It can be anticipated that the number of models applying for and receiving certification will increase significantly. However, making autonomous vehicles "ubiquitous on the roads and affordable for ordinary consumers" will not happen overnight on July 1.