Zoox Recalled 332 Robotaxis After Its Safety Logic Pushed Them Into Oncoming Traffic

What happened

In August 2025, Amazon's autonomous vehicle subsidiary Zoox issued a voluntary software recall covering its entire commercial fleet of 332 robotaxis. The vehicles had been observed crossing center lines and stopping in the path of oncoming traffic near intersections. The system was not malfunctioning in the conventional sense. It was doing precisely what it had been designed to do.

The root cause traced back to the vehicles' intersection-handling logic. Zoox had programmed the cars with what the company described as an "abundance of caution" to avoid blocking cross-traffic. That caution turned counterproductive. Near complex intersections, faulty path-planning caused the system to execute turns so poorly that it nudged the vehicles across the yellow center lines, creating head-on collision hazards that drivers in the oncoming lane had no reason to anticipate from a vehicle operating in autonomous mode.

The recall was Zoox's third significant software rollback of 2025. The company had already issued separate recalls earlier in the year for problems with pedestrian detection and unexpected hard braking. The algorithm at the center of this latest recall had been certified for public road use only months before the fault was identified. Certification had not caught it, and the vehicles had been in commercial operation during the interval between certification and recall.

Zoox addressed the defect through an over-the-air software update, pushing the fix to the fleet without withdrawing the vehicles from service. The company disclosed the defect to the National Highway Traffic Safety Administration, which logged it under Safety Recall Report 25E090. OTA remediation is now standard practice in the autonomous vehicle industry, and in this case Zoox acted proactively. But the same mechanism that makes fast fixes possible also means that safety-critical code reaches public roads during the period when its behavior under real conditions is still being learned.

The incident points to a structural gap that the NHTSA's Standing General Order was designed to address. That rule compels autonomous vehicle companies to report near-miss behaviors, and it has become one of the few tools regulators have to look inside systems whose core logic is proprietary. The fault here lived in algorithms that were not externally auditable until they produced observable errors at scale. A provable record of what a system did at each decision point, preserved before a recall is necessary rather than reconstructed after, would move accountability upstream, from the moment a car crosses a line to the moment an operator signs off on putting it in public traffic.

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

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot 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.