China tightens grip on robotaxis with new safety, liability rules

China’s 15th Five-Year Plan strategically centers technological advancement, particularly AI-driven mobility, as a national priority, catalyzing domestic champions like XPeng and Pony.ai to spearhead innovation in autonomous vehicles. By early 2026, XPeng's focus on intensive R&D, including ventures into humanoid robotics, and Pony.ai’s deployment of real-world driverless robotaxis demonstrate how these companies are not only aligning with but actively advancing government ambitions. This synergy between policy and enterprise accelerates China’s leadership in the autonomous mobility sector, supported by a regulatory environment that fosters rapid development and scaling of AI-powered transport solutions.

China’s introduction of its first mandatory national safety standard for L3 and L4 autonomous driving, GB 44721-2026, marks a decisive regulatory shift prompted by the 2025 Xiaomi fatal crash and targets models from domestic players like Deepal and Arcfox. This standard demands that autonomous vehicles perform at a safety level equivalent to a competent human driver, incorporating stringent requirements such as driver readiness monitoring for L3 systems and clear protocols for automated driving activation and deactivation. By aligning with recent UN global regulations, China is moving from flexible oversight to enforceable safety mandates that emphasize lifecycle safety management and robust risk mitigation.

The GB 44721-2026 standard enforces a comprehensive, end-to-end safety management regime requiring manufacturers to conduct simulation, test-track, and real-world road testing, alongside transparent communication to users about system limitations. A particularly challenging technical hurdle is the Minimum Risk Maneuver (MRM) requirement, which mandates autonomous systems to safely decelerate and navigate to a minimal risk condition under adverse weather or sensor failure scenarios—a task especially difficult for pure-vision systems relying solely on cameras. Additionally, Level 4 systems must operate independently without remote assistance for dynamic driving tasks, underscoring the emphasis on intrinsic system reliability, though remote cloud monitoring remains permissible for robotaxi fleets.

Perhaps most impactful for foreign automakers, China’s safety standard introduces a localized, unified certification process that is entirely independent of existing European or American approvals, creating significant market access barriers. With a hard compliance deadline set for July 1, 2027, any autonomous vehicle failing to meet GB 44721-2026 will be barred from production, sale, or import into China. This localized certification framework includes mandatory third-party evaluations based on China-specific test scenarios, effectively compelling global manufacturers to invest heavily in domestic testing infrastructure and compliance efforts.

To enhance accountability and post-incident analysis, the standard requires autonomous vehicles to be equipped with a Data Storage System for Automated Driving (DSSAD), akin to an aircraft’s black box. This mandate reflects China’s rigorous approach to safety oversight, ensuring operational data is meticulously recorded to facilitate investigations and continuous improvement in autonomous mobility safety.

China’s evolving legal framework for autonomous vehicles addresses the longstanding ambiguity surrounding liability in accidents involving self-driving technology. Traditionally, vehicle owners bore responsibility, but the rise of complex software and hardware systems complicates fault attribution, especially when vehicles operate in fully autonomous mode without active driver control. This legal uncertainty, compounded by risks such as hacking and cybersecurity breaches, has been a significant barrier to adoption, prompting China to develop clearer rules that assign liability based on the activation of autonomous functions rather than mere ownership or driver presence.

In a landmark move, China’s draft revision to the Road Traffic Safety Law introduces a dedicated chapter specifically for autonomous vehicles, distinguishing fully autonomous driving from assisted driving and assigning traffic violation liability to manufacturers or importers when autonomous systems are active. This nuanced approach stops short of automatically transferring all accident compensation liability to automakers, instead requiring proof that violations are linked to the autonomous system. Enforcement hinges on determining whether autonomous functions were engaged, though the draft leaves open questions about data access and evidentiary standards in disputes.

Complementing the legal amendments, China mandates rigorous technical standards for Level 3 and Level 4 autonomous systems, effective July 2027, requiring these systems to match or exceed the safety of attentive human drivers and to include driver monitoring capabilities for Level 3. This regulatory framework also enforces compulsory traffic accident liability insurance for autonomous vehicles and encourages commercial insurance, while imposing strict cybersecurity and data security obligations on manufacturers. These measures collectively aim to foster consumer confidence and accelerate the commercial rollout of autonomous vehicles, with dozens of new models expected to hit Chinese roads soon.

China’s centralized regulatory environment enables a proactive and tightly controlled approach to autonomous vehicle deployment, contrasting sharply with the more fragmented and reactive frameworks seen in the US and Europe. For instance, after a mass Baidu robotaxi outage in Wuhan, regulators swiftly froze new Level 4 permits for three months, demonstrating a cautious stance toward safety and public trust. Additionally, strict marketing controls prohibit misleading claims about assisted-driving technologies, exemplified by Tesla’s rebranding of its Full Self-Driving system to 'Intelligent Assisted Driving' following regulatory fines, underscoring China’s commitment to clear liability distinctions and consumer protection.

China’s updated rules require compulsory accident liability insurance and robust cybersecurity compliance for robotaxis, pushing automakers to invest in risk management and transparent data practices.

Foreign autonomous vehicle players in China face a uniquely challenging competitive landscape shaped by stringent institutional and regulatory constraints that isolate their technology development from global counterparts. Tesla’s Shanghai facility alleviates some operational hurdles, yet data localization requirements create segmented training data silos, complicating over-the-air updates and delaying advanced feature approvals. Meanwhile, mandatory national safety standards enforced by the Ministry of Industry and Information Technology demand rigorous centralized evaluations, including extensive simulations and phased real-world trials, raising the entry bar and ongoing compliance costs for all players.

Legal prohibitions on foreign entities collecting high-precision geographic data compel companies like Tesla to rely on partnerships with domestic digital map providers such as Baidu, embedding operational dependencies that affect real-time perception and navigation. This reliance, combined with the complex traffic conditions in Tier-1 cities like Shanghai and Shenzhen, forces autonomous systems to finely balance safety compliance with adapting to aggressive local driving behaviors, increasing liability exposure and operational risk.

Tesla’s competitive position in China is further complicated by the 2027 rollout of stringent L3/L4 autonomous vehicle regulations emphasizing data governance and safety accountability beyond mere technical performance. Analysts like Xing Lei of AutoXing underscore that regulatory approval hinges heavily on how Tesla manages vehicle data, not just on driving capabilities, prompting Tesla to prioritize deploying its full FSD system ahead of formal L3/L4 certification. This regulatory environment, coupled with data secrecy challenges also seen in Europe, illustrates how opaque data governance policies critically shape market access for foreign players.

Domestic players like Cao Cao navigate a volatile regulatory landscape where approval delays and abrupt permit freezes—exemplified by the March 2026 suspension following Baidu’s Wuhan robotaxi incidents—pose existential risks to scaling driverless operations. Local governments’ concerns about worker displacement give Cao Cao’s hybrid human-autonomous ride-hailing model a strategic edge, enabling a softer transition that may align better with regulatory priorities. However, the looming risk that competitor missteps could trigger market-wide pauses threatens Cao Cao’s ambitious 100,000-unit volume target by 2030, which is critical for achieving competitive unit economics below $35,000 per vehicle.

China’s evolving regulatory framework, including proposed amendments to the Road Traffic Safety Law assigning manufacturer liability for traffic violations under autonomous driving, creates a distinct liability regime that differentiates fully autonomous vehicles from those with driver-assistance features. Municipal advantages, such as Shenzhen’s pioneering liability law and Beijing’s issuance of Level 3 permits to Changan and BAIC, underscore how city-level regulatory actions significantly influence competitive positioning. This centralized, stringent approach contrasts sharply with the fragmented and comparatively lax U.S. regulatory environment, potentially favoring domestic players adept at navigating China’s tightly controlled automotive sector.