China’s autonomous driving race shifts from hype to high stakes

(By Caixin journalists An Limin and Ding Yi)

For years, China’s hyper-competitive smart electric vehicle (EV) market ran on a heady cocktail of AI breakthroughs and marketing bravado.

To match the hype Tesla Inc. created with its end-to-end automated driving system, Chinese automakers coined jargon of their own, such as “Level 2++” and “Navigate on Autopilot”, promising technology that could guide vehicles through chaotic city streets with the skills of a seasoned chauffeur.

For many consumers, the boundary between a car that assists the driver and one that actually drives itself had dissolved into a foggy haze of marketing gimmicks, sometimes with tragic real-world consequences.

Now, the era of Wild West ambiguity is coming to an end. In mid-2026, Chinese regulators introduced twin mandatory safety standards for assisted and autonomous driving systems. Scheduled to take effect next year, the standards draw a sharp, legally binding line between assisted driving and actual autonomy.

By establishing stringent testing protocols and clear liability rules, Beijing aims to eliminate consumer deception and curb safety hazards.

For automakers, meeting these benchmarks is no longer just a regulatory hurdle, it is a multibillion-RMB watershed that will distinguish deep-pocketed tech leaders from poorly funded also-rans, fundamentally rewriting the rules of commercial competition in the world’s largest automotive market.

Putting driver assistance to the test

Under China’s classification system for automated driving, Level 1 and Level 2 still count only as driver assistance, meaning a human driver must be in control of the car at all times. Yet, as automakers jostled for market share, products with wildly divergent capabilities were routinely bundled under vague, exaggerated labels.

The mandatory standard on assisted driving neatly categorises Level 1 and Level 2 features into three distinct tiers: a basic single-lane system, a basic multi-lane system and a navigation-assisted system. More importantly, it outlaws the pervasive industry practice of selling half-baked software to consumers while relying on over-the-air updates to patch safety flaws after the fact.

Under the new rules, every driver assistance system must undergo rigorous field testing and certification conducted by government-accredited third parties before mass production can begin. “In the past, safety validation was largely left to the automakers themselves,” said Liu Hui, a safety expert at SZ Zhuoyu Technology Co. Ltd., a smart car technology developer affiliated with drone giant SZ DJI Technology Co. Ltd.

Liu stressed that the new rules replace self-assessment with standardised trials that are difficult to pass. The tests leave little room for error. Vehicles will be pushed to their limits in demanding scenarios like pitch-black environments, where assisted driving systems must rely entirely on the cars’ onboard lighting, according to Liu.

Furthermore, the standard places heavy weight on driver monitoring systems and user education. Systems must continuously track a driver’s attention. If a driver ignores warnings, the vehicle must execute an automated risk-mitigation manoeuvre and safely disengage. Crucially, onboard hardware must log real-time driving data to ensure that liability can clearly be determined in the event of a serious crash.

While the requirements will drive up costs, industry insiders see them as long overdue. “Top-tier car companies have been eager for these standards, as a single catastrophic accident caused by an unregulated system can derail the entire industry,” said a person close to Huawei Technologies Co. Ltd., which doubles as a supplier of smart car tech.

Instead of competing in a race to the bottom on price, carmakers are now forced to refocus on core engineering and sensor hardware.

However, there is already a looming bottleneck: Only five testing institutions, including China Automotive Technology and Research Center Co. Ltd., are accredited to certify cars with automated driving capabilities. As automakers race to certify new models slated to hit the market in 2027, the queues at these facilities are growing longer.

Defining autonomy

While the driver assistance standard sets boundaries for human-monitored systems, the autonomous driving standard establishes a formidable bar for handing the wheel over to the machine.

To qualify as a Level 3 or Level 4 technology — both identified as autonomous driving under China’s classification framework — systems must perform as well as or better than a “qualified and focused human driver”, without posing unreasonable safety risks to others on the road.

For Level 4 systems, the bar is even higher: vehicles must independently navigate to a safe position without relying on a driver at the wheel.

Regulators also demand transparency. Automakers must explicitly inform users of system limits via in-car displays and other official channels, issuing clear indicators when the autonomous driving mode is active or disengaged.

The technical hurdle required to achieve this settles a longstanding industry debate over whether Level 3 is a necessary stepping stone or a skippable stop on the road to full autonomy. As hardware configurations for Level 3 and Level 4 systems are broadly similar, the primary differences boil down to the breadth of applications and safety redundancy, said a car company technician. “No one can guarantee an autonomous driving system will never fail,” he said. “Attempting to leap straight to Level 4 for mass deployment carries unmanageable risks.”

Level 3, he added, remains an essential proving ground.

Even so, taking Level 3 vehicles to market has proven complex. In December, Chongqing Changan Automobile Co. Ltd. and BAIC BluePark New Energy Technology Co. Ltd. gained conditional product access approval for two models with Level 3 autonomous driving capabilities. However, the vehicles were restricted to designated test roads in Chongqing and Beijing and barred from being sold at retail.

Transitioning from tightly controlled pilot programmes to commercial mass production requires passing rounds of gruelling simulation, proving ground and open road testing.

Weeding out the weak 

The dual standards have set a high bar for both carmakers’ technical expertise and the strength of their finances, accelerating the polarisation of the market. “Meeting the mandatory standards has raised the technical and investment thresholds by an order of magnitude,” said Pan Yifeng, chief engineer of intelligent driving at Changan Auto.

Pan said that it can easily cost billions of RMB to develop the software, hardware and testing infrastructure needed to build a compliant Level 3 system.

This steep cost of entry has split the market into two distinct camps. Tech giants and well-funded automakers are aggressively pressing their advantage. A new wave of approvals for Level 3 pilot programmes this year will lay the groundwork for the technology’s large-scale commercial application in 2027, said Jin Yuzhi, CEO of Shenzhen Yinwang Intelligent Technology Co. Ltd., a firm spun out of Huawei’s smart car business unit, in a media interview in July.

For Level 4, Jin said that robotaxis currently confined to demonstration zones will likely secure licences to drive on designated public roads within the year, potentially paving the way for widespread commercial deployment in 2028.

Smaller automakers, however, are stepping away from autonomous driving due to concerns that a miscalculation could lead to financial ruin. Yet staying comfortably parked in the driver assistance zone carries its own risks. The consumer market for basic driver assistance has dissolved into a margin-crushing price war and investor interest in pure Level 2 suppliers has waned due to their weakened profitability prospects.

However, the ultimate reward for those who cross the autonomous driving threshold could be a fundamental shift in monetisation models. Once Level 4 technology matures, drivers will be willing to pay for autonomous driving tech that effectively serves as a personal chauffeur, creating a new revenue engine, Li Auto Inc. President Ma Donghui said in March last year.

Ma’s view was echoed by Changan Auto’s Pan, who believes that the widespread adoption of autonomous driving services will weed out the weak and shift the industry away from self-defeating competition and toward a healthy race to create the best tech.

This article was first published by Caixin Global as “In Depth: China’s Autonomous Driving Race Shifts From Hype to High Stakes”. Caixin Global is one of the most respected sources for macroeconomic, financial and business news and information about China.

Related: Can China lead the way in autonomous driving? | Built for chaos: Why China’s robotaxis are streets ahead