Yesterday China’s market regulator and Xiaomi announced an over-the-air recall of 116,887 SU7 Standard sedans, every one built before August 30, to fix a highway assisted-driving system the regulator says shows “insufficient recognition, warning and handling ability” in extreme conditions. No parts get replaced; it is a software update. The recall is the bureaucratic epilogue to an incident that occurred on a highway in Anhui on the night of March 29.
That night, a base SU7 Standard on an expressway struck a concrete barrier in a roadwork zone where traffic had been diverted. Three young women died. It was the first fatality involving Xiaomi’s first car, and I am going to write about it carefully, because three people are dead and their families are owed sobriety rather than hot takes.
Xiaomi released a timeline pulled from the car’s own logs. The NOA system was engaged at 116 km/h, it flagged the obstacle ahead and began to slow, the driver took back manual control roughly a second before impact, and the car still hit the barrier at 97 km/h. The automatic emergency braking never fired, because the barrier was a static object the system is not designed to catch. That last sentence is the technical heart of the whole thing.
The SU7 Standard runs vision-led assisted driving: a single Nvidia Orin chip, one millimeter-wave radar, a set of cameras and ultrasonic sensors, and no LiDAR. The Pro and Max trims add a roof-mounted LiDAR and a second Orin. Long-range detection of stationary objects is a well-known weak point for camera-and-radar systems, which deliberately filter out static returns to avoid braking at every overpass and signpost. At highway speed, that leaves only a sliver of a second when something solid appears in a lane that should have been clear. This is not unique to Xiaomi; it is the same physics that keeps Tesla’s vision-only Autopilot under investigation. But Xiaomi shipped a car in barely two years and marketed the driving system hard.
The crash set off a much larger argument in China about how this technology gets sold. Regulators signaled they would start policing “intelligent driving” marketing language. Clips of drivers taking their hands off the wheel, in one case apparently asleep, circulated widely, and state-linked outlets pressed the point that these are Level 2 systems, assistance and not autonomy, and that an enormous number of drivers have been sold the opposite impression. Xiaomi’s stock shed about $15 billion in two days. Lei Jun said the company would not dodge its responsibility, which is the correct thing to say and also the only thing he could.
Which is how we arrive at yesterday’s recall. The fix is a remote update intended to improve how the system recognizes and warns in extreme conditions, and it has been pushed to every Standard built before late August. Remote recalls are routine now, but the framing is what matters here: the regulator, not just the company, put its name on the finding that the system fell short. That detail turns a single tragedy into a precedent that other Chinese automakers will now be measured against.
I am not going to adjudicate fault. The police investigation and any claims that follow are not mine to settle, and the exact braid of human and machine decisions in those final seconds may never be fully public. What I will say is that the SU7 launch sold a base car with vision-only assisted driving to buyers who, in at least some cases, trusted it more than its own makers intended. The cheapest hedge available was to put LiDAR on every trim, and Xiaomi chose the leaner bill of materials instead. The YU7 quietly extending LiDAR down its range tells you the company learned something from this.
The recall closes the regulatory loop, not the human one. A software push does not make three families whole, and the larger question, whether China’s ship-fast-and-refine-later approach to assisted driving is safe at the scale it is now reaching, is far bigger than one company and one car. Xiaomi will almost certainly sell more SU7s next month than it did before the crash. That is the part I find hardest to sit with.
Related: Five Years Inside the Xiaomi Ecosystem, my history of how Xiaomi built the machine that got it here.