L3 automation needs multilayered safety net

As a tech reporter at China Daily, I always have a front-row seat while witnessing the breathtaking pace of technological change.

Recently, I put a domestic auto model's intelligent driving assistance system to the test. The location: a notoriously chaotic intersection near Beijing's Chaoyang Joy City, a bustling hub where cars, pedestrians and a relentless swarm of delivery e-bikes converge. To my surprise, the system handled it with remarkable composure, navigating the crowding, the sudden cuts, and the complex interactions with a level of predictability that felt almost natural.

This experience was a powerful microcosm of the "intelligent driving boom" sweeping China's auto industry. We are in a transformative era, where fierce competition and rapid technological iteration are indeed making smart driving features more accessible, fueling the industry's high-quality development.

Yet, this exhilarating progress is shadowed by a pressing concern. The very competition driving innovation is also pushing some players into a vicious cycle of "involution", where exaggerated marketing often blurs the line between aspiration and reality.

As the National Development and Reform Commission rightly pointed out recently, disorderly competition and inflated claims are not just short-term profit-eroders; they pose a long-term threat to technological innovation, product safety and ultimately, the industry's global competitiveness.

The start of 2025 has seen this competition intensify, with over a dozen automakers launching new intelligent models. But beneath this fervor lie critical, unaddressed issues: immature regulations, blurred usage boundaries, a lack of public tech literacy and a muddle of marketing concepts. As Ouyang Minggao, an academician at the Chinese Academy of Sciences, cautions, we are in the era of "intelligent driving assistance for all", not "autonomous driving for all".

China's classification of driving automation standards provides a clear, six-level framework, from L0(no automation) to L5 (full automation). The stark truth is that the current practical application level for mass-market vehicles remains firmly at L2, where the driver must constantly supervise. However, in the battle for market share, we see a proliferation of misleading terms like "L2.99", "zero intervention", or "hands-free", creating dangerous misconceptions and potentially lulling drivers into a false sense of security.

Ouyang's perspective is crucial. While acknowledging that multimodal large language AI models can empower higher-level autonomy, he emphasizes that resolving their safety and reliability is paramount. Progress to L3 (conditional automation) is gated by legislation, while L4 (high automation) requires extensive, iterative realworld experience. The leap is not merely a software update.

The consensus among clear-eyed industry leaders is that we are still a considerable distance from true, widespread L3. The path is not isolated; it is intrinsically tied to the entire ecosystem — traffic infrastructure, legal frameworks and operational models. At its core, as discussions with experts reveal, is the fundamental requirement for sufficient technical redundancy. It is this multilayered safety net — in algorithms, software and hardware — that can potentially enable an L3 system to handle the infinite "corner cases" of the real world.

Companies like Didi are already implementing this philosophy. Their approach involves triple-layer redundancies to ensure safety. Zhang Bo, CEO of Didi Autonomous Driving, explained a crucial component: an independent algorithmic module dedicated solely to predicting potential collisions and initiating evasive action — a dedicated guardian angel focused only on safety.

This brings me back to my core belief, reinforced by my experience at that chaotic intersection and the broader industry landscape: For intelligent driving, safety is not just a feature; it is the ultimate luxury. It is the most sophisticated, valuable and nonnegotiable component of this technological revolution. A smooth ride through heavy traffic is impressive, but it is meaningless without an unwavering, redundant and rigorously tested foundation of safety.

Feng Xingya, chairman of GAC, hit the nail on the head: "Safety is the bottom line for auto enterprises, and this bottom line cannot be compromised." As we approach the anticipated tipping point for intelligent vehicles in the coming years, we must collectively shift our focus. The race should not be about who coins the most ambitious marketing slogan, but about who builds the most resilient, transparent and trustworthy safety systems.

For consumers, the true mark of a premium smart car will not be its ability to navigate a busy mall crossing, but the profound, quiet confidence that comes from knowing every possible failure and edge case has been accounted for. That is the luxury we should all be demanding.