The Road Traffic Safety Law Amendment Draft is currently open for public comment. This law, which has been in effect since 2003 and has undergone three revisions in 2007, 2011, and 2021, with its total number of articles consistently maintained at 124, will now see its provisions expand directly to 170 articles. It introduces a special chapter on "Autonomous Vehicle Special Provisions," with nine entirely new articles spanning Articles 95 to 103, marking the first time autonomous driving is incorporated into the road traffic management system at the national legal level. The scale of this revision far exceeds the previous three amendments. Among the provisions, the one with the most significant impact on the insurance industry is Article 97: When an autonomous vehicle commits a road traffic violation while its autonomous driving function is activated, the autonomous vehicle manufacturer or importer shall accept the penalty; if the manufacturer or importer believes the violation is unrelated to the autonomous driving function, it shall bear the burden of proof.
Shang Yanyan, President of Cheche Technology, stated in an interview that previously, some smart driving feature advertisements on the market easily misled consumers into believing that after activating assisted driving, they could completely hand over the driving task to the system. In reality, assisted driving does not mean the driver can abandon observation, monitoring, and the necessary obligation to take over. The law clearly defining relevant boundaries helps clarify the respective responsibilities of drivers and technical systems, reducing disputes over liable parties after accidents occur. "For the insurance industry, this also means that in the future, determining liability and handling claims cannot rely solely on whether the accident occurred while the vehicle's assisted driving function was activated. It also requires determining what function was enabled at the time, whether the system was operating normally, whether the driver fulfilled their due obligations, and what exactly caused the accident," Shang Yanyan said.
Conceptual Boundaries Clarified at the Legal Level for the First Time
Article 95 of the amendment draft defines two core concepts: an autonomous vehicle refers to a vehicle that, under designed operating conditions, continuously and automatically performs all dynamic driving tasks in place of the driver; an assisted driving function refers to a function that, under designed operating conditions, assists the driver in performing part of the dynamic driving tasks. Vehicles with autonomous driving functions not activated, or vehicles equipped only with assisted driving functions, continue to follow traditional motor vehicle management rules. This distinction may appear to be a definitional issue, but it is actually the logical starting point for subsequent liability allocation and insurance rule-making. In the past, the boundary between "autonomous driving" and "assisted driving" has long been blurred both inside and outside the industry, with many promotional pitches packaging assisted driving as autonomous driving, leading to continuous disputes over liability attribution after accidents. The draft uses "whether all dynamic driving tasks are continuously and automatically performed" as the dividing line, explicitly excluding assisted driving from the autonomous driving special chapter, which amounts to confirming at the legal level: under assisted driving status, the driver remains the liable party, and traditional auto insurance rules continue to apply; only when the autonomous driving function is activated does the liable party shift.
Shang Yanyan pointed out that in traditional driving scenarios, the driver is usually the direct controller of the vehicle, and accident liability determination is relatively clear. However, in autonomous driving scenarios, automakers hold key data such as vehicle operation logs, system decision records, and takeover prompts, naturally giving them a stronger information advantage. Therefore, requiring enterprises to bear the burden of proof under specific circumstances is essentially aimed at reducing the information asymmetry between consumers and technology companies through adjustments to liability and evidentiary mechanisms, making liability determination fairer and also pushing enterprises to improve system safety and data management capabilities. "If no one steps up when violations occur under activated status, traffic rules become meaningless; there must be an entity to accept the penalty," Jiang Han, a senior researcher at Pangoal Institution, told our reporter. The underlying data and logs are all in the hands of automakers; it is nearly impossible for car owners to prove that it was a system problem. Reversing the burden of proof essentially turns data obligations into corporate obligations. Moreover, some companies previously packaged assisted driving as autonomous driving and shifted risks to users. Now, exaggerated advertising will lead to liability, forcing them to genuinely focus on algorithm safety and data traceability, so the industry can return from traffic-driven competition to deep technological cultivation.
The allocation of the burden of proof is another key aspect of Article 97. If an automaker claims that a violation is unrelated to the autonomous driving function, it must provide evidence itself. Automakers that control underlying operational data and system logs thus transform from "data holders" to "parties with evidentiary obligations." Public security traffic management departments can retrieve relevant vehicle data for investigation, providing a superior legal basis for liability determination.
Shift in Liable Parties Forces Reconstruction of Product Systems
When the liable party changes, the design logic of insurance products must inevitably follow. Under activated autonomous driving status, the liable party shifts from "the person" to "the manufacturer or importer," and one way for automakers to transfer this liability is to purchase corresponding liability insurance from insurance companies. Cui Dongshu, head of the China Automobile Dealers Association Passenger Car Market Information Joint Committee, has publicly stated that after the draft takes effect, the future auto insurance market may form a pattern of two products operating in parallel: one is "non-activated status" auto insurance purchased by car owners, covering manual driving scenarios; the other is "autonomous driving product liability insurance" purchased by automakers, corresponding to risk liability under activated status. For insurance companies, the challenge is by no means just adding a new type of coverage, but rather rebuilding the product system, pricing models, and claims processes.
The changes in the claims investigation process are the most intuitive. In the past, auto insurance investigations mainly looked at collision marks, on-site positions, and driver statements, but under activated autonomous driving status, insurance companies need to retrieve driving data, system logs, takeover records, and other information to determine what operating state the vehicle was in when the accident occurred. "The challenge autonomous driving poses to the insurance industry is not just adding new insurance liabilities, but rather that traditional auto insurance risk identification and pricing logic need upgrading," Zhu Junsheng, postdoctoral fellow in applied economics and professor at Peking University, told our reporter. The main shortcomings at present are threefold: first, insufficient accumulation of actual operational data makes it difficult to accurately assess risks across different technologies and scenarios; second, the ability to identify and price technical risks such as algorithm, software, and system failures needs improvement; third, complex accident investigation, data acquisition, and cross-party subrogation mechanisms still need refinement.
In Zhu Junsheng's view, in the future, commercial auto insurance will not simply shift from "insuring people" to "insuring systems," but will gradually form a multi-layered protection system covering drivers, vehicles, autonomous driving systems, and related corporate liabilities. Specifically, insurance products will place greater emphasis on driving modes, system status, and operating scenarios, and coverage will further connect with mechanisms such as product liability insurance. At the same time, insurance services are expected to extend from post-accident compensation to risk prevention and safety management.
Shang Yanyan believes that the most realistic shortcoming of the insurance industry today is not a lack of new insurance products, but rather the ability to identify, price, and judge liability for smart driving risks. She further pointed out that traditional auto insurance pricing mainly relies on factors such as vehicle model, driver's historical accident records, and vehicle usage. But in the era of smart driving, the same car model may be equipped with different versions of software systems, and different system versions may differ in perception capabilities, decision logic, and applicable scenarios. Even for the same vehicle, whether the system is activated, whether it exceeds designed operating conditions, and whether the driver takes over in time can all affect the cause of the accident and liability allocation. "This places new demands on insurance companies," Shang Yanyan said. First, they must possess stronger data acquisition and analysis capabilities to understand vehicle operating status, system versions, and accident records; second, they must establish risk pricing models oriented toward smart driving scenarios, rather than relying solely on traditional vehicle model and driver dimensions; third, they must enhance liability analysis and claims coordination capabilities for complex accidents, able to distinguish driver fault, product defects, and the responsibilities of other relevant parties.
Comments