Summary of Key Points
Li Jun, an academician from the Chinese Academy of Engineering, stated at the 2026 China Auto Forum that the autonomous driving industry will completely move away from the "driver-full responsibility" model and enter an era of operational safety. The main changes include the shift of responsibility from drivers to automobile manufacturers, as well as an upgrade in safety requirements from a one-time conformity upon factory release to continuous safety throughout the entire lifecycle of the vehicle. Automobile manufacturers will be required to prove their vehicles' safety at all times, monitor them in real-time while they are on the road, and assume lifelong responsibility for any issues that arise. Additionally, the mandatory national standard GB 44721 will come into effect in 2027, leaving automobile manufacturers with only one year to adapt their systems. The industry faces multiple challenges, such as a reversal of responsibility, the need to restructure processes, and overcoming technical bottlenecks.
1. Major Reversal in Liability Logic: In the Event of an Autonomous Driving Accident, Automobile Manufacturers Bear the Initial Blame
In the past, when an accident occurred while driving, the focus was on the driver's actions. Now, in the event of an autonomous driving accident, regulatory authorities, insurance companies, and courts will immediately hold the automobile manufacturers accountable, examining whether there are any bugs in the algorithms, whether sufficient data has been retained, and whether there are flaws in the safety control processes.
More importantly:
- The certification process has changed: In the past, obtaining qualifications required a few tests in a laboratory or on simulated roads. Now, a "Safety Case" must be presented to demonstrate the safety of the system. This means providing a comprehensive logical chain of evidence, such as technical principles, test data, and scenarios that show the vehicle's safety at a speed limit of 60 km/h in urban areas.
- The regulatory oversight has become more stringent: Instead of simply ensuring conformity upon factory release, vehicles will be subject to lifelong monitoring, with all driving data being retained as evidence of safety.
2. Three Major Challenges for Automobile Manufacturers within One Year
Automobile manufacturers have only one year to adapt to the new regulations, facing three main pressures:
1. Process Reengineering: Traditional development processes involve three steps: design, testing, and market launch. Now, a fourth step—full-life-cycle monitoring of vehicle operations in the market—is necessary. This means that manufacturers must continuously monitor the autonomous driving systems of each vehicle after sale and promptly fix any issues through over-the-air (OTA) updates.
2. Building New Systems from Scratch: Manufacturers need to establish two new systems:
- A legal framework to prove their innocence in case of an accident, including the ability to quickly determine if the vehicle was safe at the time of the incident, regularly self-inspect for compliance, and have evidence to prove no fault.
- An online monitoring system to continuously monitor vehicles on the road, identify dangerous situations (such as heavy rain or construction), retain extensive data, and promptly identify and address potential issues.
3. Time Pressure: The new standard will take effect in July 2027, and traditional manufacturers, lacking experience, face significant challenges in developing these two systems from scratch.
3. Four Core Challenges: The Battleground for Technological Competition Among Automobile Manufacturers
The new regulations address four critical issues that will be essential for the future success of automobile manufacturers:
1. How to Measure Continuous Safety: There is no unified standard for determining what level of accident and failure rates constitutes "safety."
2. How to Ensure Algorithm Trustworthiness: The decision-making logic of AI (e.g., why a sudden brake application occurred) and visual perception (e.g., whether pedestrians were detected) must be fully verifiable and traceable.
3. Acceptance of Residual Risks: Autonomous driving cannot eliminate all risks; how can the public and regulators accept the remaining small probabilities of accidents, especially under extreme weather conditions?
4. How to Handle Unknown Situations: How can AI systems learn from unseen scenarios (e.g., suddenly appearing animals) and continuously improve their performance?
To address these challenges, the academician's team has proposed five tools, two of which are mandatory:
- Safety Case: A detailed document that explains under what conditions the vehicle can operate safely, provides evidence of its safety capabilities, and highlights any limitations, to reassure regulators and consumers.
- ISMR (In-Service Monitoring System): A system that monitors all vehicles in real-time, identifies dangerous situations, uploads data for analysis, and performs OTA updates to fix issues, effectively equipping each vehicle with a "safety monitor."
4. Unovercome Technical Bottlenecks: The Industry Is Still Navigating in the Dark
Although the regulatory framework is in place, four key issues remain unresolved:
1. Lack of Clear Standards for Continuous Safety: There is no consensus on what level of accident rates constitutes "continuous safety."
2. Difficulty in Defining the Scope of Autonomous Driving (ODD): While autonomous vehicles have designated safe driving areas (e.g., they cannot operate on rural roads), there is no precise method to determine when a vehicle has left these limits.
3. No Monitoring for System Degradation: Over time, sensors and chassis components may degrade, affecting the performance of AI algorithms, but there are no mature monitoring methods to address this issue.
4. Unclear Thresholds for Residual Risks: There is no consensus on what level of risk is considered "acceptable," with only Japan having preliminary guidelines.
5. Impact on the Industry: Shifting from Selling Vehicles to Providing Services
Under the new regulations, automobile manufacturers will be responsible for the safety of their vehicles throughout their entire lifecycle, shifting the focus of research and development from simply building good cars to managing them effectively. This may lead to changes in business models, with manufacturers potentially earning revenue through subscription-based autonomous driving services rather than just selling hardware. Smaller companies may struggle due to the high cost of implementing these new systems.
Academician Li Jun concluded by calling for the industry to quickly establish unified safety standards to provide a clear "safety baseline" for the development of autonomous driving, ultimately prioritizing safety above all else.
The core message of this news is that the safety rules for autonomous driving have undergone a fundamental change, shifting the responsibility from manufacturers to ensuring the safety of vehicles. This requires a comprehensive reengineering of the industry's research and development, production, and after-sales systems—a revolution in both technology and business models.