On September 19, at the 2026 TEDA Automotive Forum, Xiao Lingyun, Director of the Industrial Products Department of the Defective Product Administrative Center of the State Administration for Market Regulation (SAMR), took the stage and presented a set of figures: by the end of 2025, China had cumulatively implemented 121 million vehicle recalls, 53% of which were triggered by defect investigations initiated by regulatory authorities. By June 2026, the national vehicle ownership reached 370 million, meaning nearly one in every three vehicles had experienced a recall.
This number sounds substantial. But immediately after, he cited another figure: from 2019 to 2025, the fire rate per 10,000 New Energy Vehicles (NEVs) dropped by 68%.
More recalls, fewer fires. Placing these two sets of data together indicates that safety supervision over the past few years has been effective.
In reality, however, the true challenge has just begun. The new issues brought about by Intelligent Connected Vehicles (ICVs) have exceeded the coverage of the traditional recall system.
Nearly 8 Million Vehicles Recalled in a Single Day, All "Caused" by New Technologies
Xiao Lingyun gave an example. On August 21 this year, the Chinese market saw nearly 8 million vehicle recalls in a single day. Most of these recalls were not due to traditional mechanical failures, but were related to new technologies: software issues, battery system problems, and intelligent driving function defects, among others.
The traditional recall system was established in the era of internal combustion engine vehicles. The Product Quality Law defines "defects," and the Regulations on the Administration of Recall of Defective Automotive Products stipulate the recall process. This system has worked well over the past decade. However, new problems are now emerging one after another.
With battery swap models becoming increasingly prevalent, the regulations only govern original equipment manufacturers (OEMs); who is responsible for the safety of swap batteries? Autonomous driving systems are increasingly concentrated in the hands of a few suppliers, leaving OEMs with very limited understanding of the internal systems. Can suppliers be held as the responsible entities? Cybersecurity, data security, and privacy security—issues that did not exist in the traditional automotive era—are now occurring on a daily basis.
The regulations are currently under revision. Xiao Lingyun revealed that the new recall regulations will include motorcycles and will also address the new regulatory requirements brought by ICVs. However, the revision cycle is long, and there is no time to wait. Last year, the SAMR and the Ministry of Industry and Information Technology (MIIT) jointly issued a document, putting forward a series of requirements regarding accident reporting, sandbox supervision, and other aspects. Xiao Lingyun described this document as a milestone in the field of ICV safety supervision. The logic of traditional recalls is: an accident occurs, consumers complain, regulatory authorities investigate, and the enterprise recalls. This process has an inherent lag. By the time an accident happens, people may already be injured.
Sandbox monitoring has changed this logic. A sandbox refers to a specially built extreme stress testing environment outside of real user usage scenarios. Vehicles are placed in this environment and subjected to the most extreme operating conditions to "torture" them, proactively identifying issues before accidents occur.
Xiao Lingyun provided the data: currently, multiple vehicles have had issues discovered through sandbox testing. Enterprises have proactively made improvements based on this, preventing these problematic models from entering the market and causing more accidents and casualties. New smart models from mainstream automakers are successively participating in sandbox testing.
Behind this is a "three-line theory." Xiao Lingyun explained that the sandbox is the high line, proactively identifying risks before accidents occur; the recall is the bottom line, which must be executed once problems are exposed. In the middle is another line, which is daily accident reporting and monitoring. These three lines together constitute the safety supervision system for ICVs.
Last year, the Defective Product Administrative Center also introduced an accident reporting system. This system requires enterprises to report on time after the occurrence of intelligent driving accidents. Regulatory authorities then aggregate and analyze the data to form a dangerous scenario database.
Xiao Lingyun said that before the establishment of the accident reporting system, the industry assumed that the issues with advanced driver-assistance systems (ADAS) had been largely resolved. However, after the actual data came in, it was found that vehicles still had deficiencies in recognizing and handling irregular objects and special targets. The reason is simple: laboratory tests are based on standard scenarios, but drivers face non-standard scenarios every day.
Using AI and Data Closed-Loop to Keep Pace with Technological Iteration
The quantity and growth rate of risks associated with ICVs far exceed the processing capacity of manual analysis in the traditional automotive era. What Xiao Lingyun's team is doing is using large models and data closed-loops to elevate regulatory efficiency.
They are applying for two national standards. One is the defect determination standard for ADAS, used to determine whether L2 and L3 production vehicles have design defects. The other is a standard related to accident reporting, which standardizes the data format and content reported by enterprises.
More cutting-edge work involves using AI (Artificial Intelligence) to build product analysis tools. Regulatory authorities have already integrated the data interfaces between the accident reporting system and the dangerous scenario database, enabling the system to automatically generate new test scenarios based on historical accident data.
Currently, seven or eight enterprises are already accepting the challenges of these automatically generated scenarios in the sandbox. Xiao Lingyun said that the goal of this system is to enable quality supervision to achieve rapid identification and discovery of risks in ICVs.
Standardization efforts are also catching up. The SAMR, in collaboration with Contemporary Amperex Technology Co., Limited (CATL), successively released four national standards for power battery quality management last year. These standards were not written out of thin air; rather, they distilled the excellent management experience of leading enterprises into national standards. The next step is to transform these standards into international standards to gain a voice in global rule-making.
Xiao Lingyun said that while innovation in China's automotive industry is rapid, many innovations have also failed. Some innovations are not accepted by consumers, and safety issues have even caused consumer resistance. His judgment is that after the market washes away the impurities, what remains will undoubtedly be brands and models that excel in both quality and safety.
This judgment is far from redundant in the current market environment. Since 2026, the industry has seen fierce price wars, with many enterprises rushing to launch new cars and compete on configurations, which can easily squeeze safety investments. The role of regulatory authorities is to hold the bottom line in such times.
In addition to domestic supervision, the Defective Product Administrative Center is also building an overseas compliance platform. As China's automotive export volume grows, overseas data cannot be brought back, and enterprises are unfamiliar with the quality guarantee regulations and recall rules of target markets, making them vulnerable to losses. The platform integrates regulatory requirements and data models from major overseas markets, helping enterprises consider compliance issues during the product design phase.
Xiao Lingyun concluded by saying that safety truly defines the automobile. Only safe products can sustain consumption in the market. This statement sounds simple, but in 2026, when intelligent driving is advancing at a breakneck pace, it carries significant weight.
--END--