From a flood of user complaints to a firm rule: the mandatory L2 standard draws a clear, unified baseline

On June 27, 2026, the State Administration for Market Regulation and the Standardization Administration of China approved and published the “Safety Requirements for Intelligent Connected Vehicle Combined Driver Assistance Systems” (GB 47955—2026). This is China’s first mandatory national standard for Level 2 combined driver assistance systems and will officially take effect on January 1, 2027. Public data from the Ministry of Industry and Information Technology shows that by 2026, the penetration rate of new passenger vehicles equipped with combined driver assistance functions in China had already reached 70%, with the penetration rate for models featuring Navigate on Autopilot (NOA) exceeding 30%. Assisted driving has long since transformed from an exclusive feature on highend models into a standard function on mainstream new cars.

Alongside this trend, user complaints related to intelligent driving are growing rapidly, and disputes over feature experience and safety boundaries continue to simmer. Both complaint data from Car Quality Network and industry research by Carresearch indicate that the focus of consumer concern about vehicle quality is gradually shifting from traditional mechanical failures to the experience and safety of intelligent features.

Adoption and complaints are both climbing — time for a quality upgrade

The rapid expansion of China’s new energy vehicle market has accelerated the adoption of intelligent driving features. In recent years, automakers have routinely made assisted driving a core selling point, expanding from highway ACC+LCC combinations to navigationbased driver assistance that covers urban roads. The functional coverage has steadily broadened and the adoption threshold has continued to drop. However, the pace of feature proliferation has not kept up with the refinement of quality and safety, and the associated user complaints and grievances have been accumulating year by year.

According to fullyear 2025 complaint data released by Car Quality Network, the platform received a total of 227,800 valid complaints, an increase of 31.6% yearoveryear. Among them, complaints about driver assistance system malfunctions surged by 77.7%, far outpacing the overall complaint growth rate. In the second quarter of 2025, driver assistance system faults entered the platform’s Top 20 complaint categories for the first time, becoming a major new driver of newvehicle quality complaints. The “2025 China Automotive Product Quality Performance Study (AQR),” jointly released by Carresearch and Car Quality Network, also noted that from January to August 2025, quality and service problems with intelligent features such as invehicle connectivity and driver assistance ranked among the top 20 complaint points, and that consumer concern is shifting from “range anxiety” toward “intelligent experience anxiety.”

Looking at specific complaints, erroneous system responses are the most frequently reported user grievance. Beyond poor decisionmaking, defects in the function’s disengagement mechanism are another major source of complaints. Many owners report that the assisted driving system disengages without any warning when it goes beyond its applicable scenarios, leaving the driver no time to react. Another prominent category of complaints involves discrepancies between advertised features and the actual vehicle. Many automakers heavily promote advanced intelligent driving and fullscenario NOA capabilities at a vehicle’s launch, but after owners take delivery, those features are not delivered via OTA updates for a long time—some models even exhibit “hardware preinstallation but locked algorithms.” Customers pay for highend configurations yet are unable to use the advertised functions for an extended period, and this type of servicerelated complaint accounted for a continuously rising share of all intelligent driving complaints in 2025.

This shift in quality pain points is also directly linked to a broader change in the automotive quality landscape. According to the AQR study by Carresearch, the number of physical quality issues in automotive products declined yearoveryear in 2025, while problems related to perceived quality and experience quality rose significantly. Simply put, traditional hardware failures have become less common, but user dissatisfaction caused by software functions and interaction experience has increased. This change is especially pronounced in new energy vehicles, where experience issues with smart cockpits and driving systems have become a new flashpoint.

At the same time, users’ tolerance for new technology is declining. In the past, consumers were willing to give new features room for improvement; now they demand that features work reliably right out of the box. This shift in mindset makes quality issues with assisted driving more likely to be amplified, which in turn erodes users’ trust in the brand.

A mandatory baseline that takes direct aim at the industry’s pain points

The introduction of the mandatory national standard GB 47955—2026 is essentially about drawing a safety red line for a rapidly expanding assisted driving market. Before this, China only had recommended standards for singlelane and multilane driver assistance, with no mandatory constraints. Automakers varied enormously in system capability, safety strategies, and interaction logic, which left ample room for exaggerated claims and feature inflation. This mandatory standard directly targets the widespread pain points in the industry, laying out explicit, hard requirements across multiple dimensions—including system classification, performance requirements, humanmachine interaction, and test validation.

The national standard first establishes a clear threetier classification for combined driver assistance systems: basic singlelane, basic multilane, and navigationbased combined driver assistance. Each category corresponds to different applicable roads and functional boundaries. This clear classification directly responds to the industry’s previous problem of vague marketing. In the past, many automakers packaged their products using selfcoined terms like “L2.5” and “L2.9,” blurring functional boundaries and making it difficult for consumers to judge the systems’ actual capabilities. Once the standard takes effect, the assisted driving functions of all vehicle models will have to fall into explicit categories and label their scope of applicability, squeezing the room for marketing hype out of the rulebook.

To address the heavily complainedabout issues of handsoff driving and ambiguous humanmachine responsibility, the standard mandates that all levels of combined driver assistance systems be equipped with a driver monitoring system that cannot be turned off. It explicitly requires that the system issue an alert if the driver’s hands are off the steering wheel for 5 seconds. If the driver has not retaken control within 10 seconds after the alert, the system must escalate to highintensity audible and visual warnings. If the driver’s gaze deviates from the road ahead for an extended period, the system must also issue graded reminders over time, culminating in an immediate takeover request. For drivers who repeatedly violate these rules, the system can lock out the assisted driving function for at least 30 minutes.

The standard also explicitly prohibits automakers from implementing any feature that suppresses “hands on steering wheel” alerts, ensuring that the driver stays in primary control at all times. This rule directly puts an end to the practices of some brands that downplayed driver responsibility and tacitly allowed hands-free driving, reinforcing the driver’s primary responsibility at both the hardware and logic levels.

To tackle pain points such as inaccurate recognition and frequent erroneous responses, the national standard sets clear, hard performance metrics based on the characteristics of road traffic in China, and incorporates a large number of localized test scenarios. The standard requires that a vehicle traveling at 80 km/h be able to detect a stationary obstacle 120 meters ahead and execute a smooth deceleration to a stop. In the closed-course testing phase, the standard includes scenarios with distinctly Chinese characteristics—such as avoiding construction zones, overturned vehicles, cardboard box obstacles, and scooter-style motorcycles—none of which are mandatory in the UN R171 regulation.

In the past, many assisted driving systems performed well under normal road conditions but tended to fail when encountering construction zones and non-standard obstacles common on Chinese roads. The newly added localized test scenarios are designed to force automakers to optimize their algorithms for real-world road conditions, thereby reducing issues like phantom braking and recognition failures.

To address the safety hazard of sudden system disengagement, the national standard explicitly requires that the system incorporate graduated risk mitigation and fault degradation logic. When the vehicle leaves the designated operational road range, when perception hardware fails, or when it exceeds its operating conditions, the system must not abruptly cut off the function. Instead, it must decelerate smoothly, degrade in steps, and ultimately bring the vehicle to a safe stop.

In other words, even if a system failure occurs, the vehicle must leave the driver with enough time to react and disengage in a controlled manner, rather than simply “dumping the responsibility” onto the driver. This requirement directly addresses the user complaint pain point of “disengagement without warning” and, at the regulatory level, reduces the safety risks posed by system failures.

Beyond that, the standard also sets unified requirements for data recording, user notification, and information security. It mandates that vehicles must fully record raw data—including vehicle status, system operations, and the road environment—throughout the entire period the assisted driving function is active, and it standardizes both the storage format and the data retrieval protocol so that it dovetails with the existing EDR (Event Data Recorder) system.

This provision solves the problems that previously plagued assisted driving accidents—difficulty retrieving data and unclear attribution of liability—by providing a statutory technical basis for post-incident tracing and regulatory audits. At the same time, the standard requires automakers to clearly inform users of a system’s functional limitations before use, listing the specific scenarios the system cannot handle. Furthermore, every time the vehicle is powered on, it must confirm that the driver has completed the necessary training before the assisted driving function can be activated.

Compared with the UN R171 international regulation issued by the United Nations in 2024, this national standard remains aligned at the core framework level, but its details are far better tailored to the realities of the Chinese market. The international regulation provides only a basic functional framework, whereas the national standard makes finer system classifications, imposes stricter driver monitoring requirements, uses more localized test scenarios, and employs testing methods that are far more practical. In a sense, UN R171 sets up a universal rulebook for global use, while GB 47955—2026 fills in the specific requirements adapted to Chinese road conditions, establishing a higher safety floor and delivering stronger enforceability.

In the short term, the release of this mandatory national standard will impose cost pressure on some automakers as they adapt to the new requirements. Making driver monitoring a standard feature across all models, adding localized scenario tests, and improving data recording systems will all drive up R&D and validation costs. Products that rely on stacking hardware at low prices while neglecting safety refinement will face compliance pressure. In the long run, however, a unified safety baseline will reduce disorderly competition in the industry, correct consumers’ misconceptions about driver assistance, and ultimately lower the risks users face. Getting the standard on the books is only the first step; subsequent oversight of market access, the rigor of manufacturers’ enforcement, and the cultivation of user habits will all determine how effective it really is.