Topics Price war Tariffs Solid state ADAS Software Sales data Supply chain Battery recycling

Home /Market & Policy

Market & Policy 3 min read

China's autonomous driving liability draft puts burden on carmakers, raising data fairness concerns

A Chinese draft regulation would hold automakers responsible for traffic violations by autonomous vehicles, but evidence would rely on data supplied by the carmakers themselves, sparking debate over fairness and legal clarity.

A draft regulation in China proposes that automakers bear responsibility for traffic violations committed by their autonomous vehicles, but the evidence for enforcement would come from data logs supplied by the carmakers themselves, a setup critics say is like letting students grade their own exams.

According to a report by the Chinese outlet Huxiu, the draft, which was reported on August 25, 2026, by Jiemian News, assigns liability for traffic violations during autonomous driving to the vehicle's manufacturer or importer, while also requiring companies to ensure road safety, cybersecurity, and data security. However, the key evidence—vehicle operation logs, known as DSSAD (autonomous driving data recording system)—would be provided by the same companies being penalized, creating a conflict of interest.

Technical standards exist, but legal gaps remain

The technical foundation is not entirely new. The national standard GB 44497-2024, which mandates DSSAD for L3 vehicles starting January 1, 2026, requires recording five categories of core data: basic vehicle and system information, vehicle status and dynamics, autonomous driving system operation, driving environment, and driver operation and status. The data is classified into A-level (mandatory) and B-level (conditional). An analysis by Youjia called this a "third-party evidence" source, noting that in past assisted-driving accidents, car owners and manufacturers often disputed whether the driver or the system was at fault, and the data recording system could provide clarity.

Other standards include GB 44721-2026, which sets safety requirements for L3/L4 systems, and GA/T 2388-2026, which establishes unified driving safety rules. Together, they aim to make behavior regulatable, data traceable, and responsibility definable. However, the draft does not specify what data carmakers must provide, in what format, or within what timeframe, nor does it address incomplete data.

A case from June 2026 in Beijing's Yizhuang area highlighted the issue. A driverless vehicle operated by Pony.ai changed lanes and was rear-ended; traffic police ruled the autonomous vehicle fully at fault, but the determination relied heavily on data supplied by Pony.ai, which the other driver could not access or independently verify.

Consumers face uphill battles; experts propose shifting burden of proof

Consumers are at a disadvantage, as they lack the technical and financial means to conduct third-party assessments. Proving a system defect is extremely difficult, and legal proceedings can drag on for months or years. Additionally, over-the-air (OTA) updates allow carmakers to remotely adjust autonomous driving logic and safety thresholds without owner consent or notification. As one analysis pointed out, "a stable strategy today could become a completely different logic after an update."

Legal scholars suggest a solution. An article by Wang Junrui, published in Study Times on August 21, 2026, argues that victims should only need to prove that an accident occurred and that the vehicle was in autonomous mode, while the burden of proof for system defects and data completeness should shift to the carmaker. If a carmaker cannot prove its system was not at fault, it should be presumed liable.

Privacy and data retention also pose challenges. DSSAD records environmental perception and driver operation data, which may involve personal privacy, and law enforcement access should perhaps require judicial review. Algorithm audits involve trade secrets, balancing public scrutiny with confidentiality. The standard requires storage for at least five collision events and 2,500 timestamp events, but if an accident investigation starts after data has been overwritten, legal obligations for longer retention may be needed.

The draft leaves details to be formulated by the Ministry of Public Security, but core issues—the legal nature of data provision obligations, rules for incomplete data, and the balance between privacy and enforcement—remain unresolved. As Huxiu notes, letting carmakers supply evidence for their own liability makes enforcement hollow.