Zoox Recalled 270 Robotaxis After One Crashed Because It Predicted the Wrong Move
What happened
In April 2025, an unoccupied Zoox robotaxi struck a passenger vehicle in Las Vegas. No injuries were reported, and both vehicles sustained only minor damage. But the collision was serious enough to trigger a voluntary software recall covering 270 vehicles and a temporary suspension of Zoox's driverless operations across its Las Vegas fleet.
The sequence that produced the crash was narrow and specific. A passenger car was approaching from the side, moving slowly and perpendicular to the robotaxi's path, apparently preparing to merge or cross. The robotaxi's software read that movement as a prelude to the car continuing forward, so it slowed and steered right to give way. The car, however, stopped and yielded, staying put in the shoulder lane. The robotaxi, having already committed to that evasive line, could not brake hard enough to avoid contact.
What failed was a predictive model, not a sensor. The robotaxi could see the car. What it could not do was correctly account for a vehicle that approached slowly from the side and then stopped rather than continuing through. According to Zoox's own recall documentation, the software misjudged the behavior of vehicles slowly approaching perpendicularly and stopping, which produced an inaccurate trajectory prediction and limited the system's ability to avoid a collision in that specific scenario.
The regulatory and operational fallout was swift. Zoox issued a Part 573 Safety Recall Report to the National Highway Traffic Safety Administration, suspended driverless operations for a full safety review, and pushed a fleet-wide software update to address the flaw. The voluntary nature of the recall indicated Zoox identified and disclosed the issue itself rather than responding to regulatory demand. The episode added to a documented pattern of cases where robotaxi systems, including those operated by Waymo and Cruise, have struggled with the irregular behavior of human drivers and the unpredictability of real urban traffic.
The crash also points to a gap in how autonomous vehicle incidents get examined after the fact. A voluntary recall report describes what went wrong and what was changed, but it does not produce a continuous, independently verifiable record of how the system was reasoning in the moments before impact. Without that, every investigation begins from the same position: sensor logs, a press release, and a regulator reading a manufacturer's own account of its own failure. A provable record of what a system did, at what confidence level, and on what prediction, would make those accounts checkable rather than simply credible on the surface.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.