Zoox recalls 105 robotaxis after one entered a smoke-covered fire scene

Why it matters

Zoox controls the vehicle, software and fleet, so it can fix all 105 robotaxis centrally. That same ownership leaves Levinson's team fully accountable when the driving system misreads an emergency.

Jesse Levinson, Zoox's co-founder and chief technology officer, is confronting a weakness in the autonomous-driving system he has spent more than a decade building: heavy smoke obscured an active fire scene, and one of Zoox's driverless vehicles entered it.

Zoox has recalled automated-driving software installed on 105 company-owned robotaxis after the June 20th incident, according to a July 8th filing with the National Highway Traffic Safety Administration. The unoccupied vehicle braked hard while trying to steer away, stopped, and then reversed under guidance from a human teleguidance tactician. First responders subsequently placed cones across two of the road's three through lanes.

No injuries were identified in Zoox's analysis. Zoox told NHTSA that its fleet review found no other incident of this kind, a conclusion reported by the manufacturer rather than an independent finding. Engadget reported the recall on July 17th.

The incident cuts directly into Levinson's founding thesis. A Princeton graduate with a computer science PhD from Stanford, Levinson developed algorithms for Stanford's winning vehicle in the 2007 DARPA Urban Challenge and later led autonomous-driving research at the university. He co-founded Zoox in 2014 to build the vehicle, driving software and ride service together, rather than retrofit autonomy onto a conventional passenger car.

That integrated approach gives Zoox control over the remedy. It also makes Zoox responsible for each layer when the system encounters a road condition it cannot interpret.

Smoke exposed a perception and response gap

The affected Zoox software had been installed on robotaxis operating on public roads between April 23rd and July 8th. NHTSA's filing says the software may fail to detect and respond appropriately to heavy smoke, particularly around an active emergency. Reduced visibility can increase the risk of a crash or impede first responders.

Zoox investigated the incident from June 22nd through July 8th while developing an update that improves its existing ability to identify and react to heavy smoke. Zoox also added an interim measure intended to keep its vehicles away from active fire scenes. The Zoox Safety Committee voted on July 7th to conduct the voluntary recall.

The remedy is software rather than replacement hardware. Zoox owns, operates and directly controls all 105 affected vehicles, so there are no retail owners or dealers to notify. That structure turns a process associated with mailed notices and repair appointments into a centrally managed fleet update.

It also strips away a common excuse. Zoox does not have to persuade customers to bring vehicles into a service center or wait for dealers to install a fix. The test is whether the updated driving system handles the next obscured emergency scene correctly.

The fire scene had not yet been cordoned off with cones when the robotaxi arrived. That detail matters because autonomous systems cannot rely on a perfectly marked road. Emergency scenes change by the second. Smoke, stopped vehicles, people in the roadway and improvised instructions can precede the cones and lane closures that make the situation easier for a machine to classify.

Regulators are treating emergency scenes as routine driving

Zoox filed the recall on the same day NHTSA issued a broader warning to autonomous-vehicle developers. The agency said it had documented driverless vehicles entering active emergency scenes, blocking ambulances and firefighters, or failing to respond to flashing lights, flares, smoke, fire and traffic cones.

NHTSA Administrator Jonathan Morrison called that behavior a "functional insufficiency" and instructed developers to prioritize first-responder interactions. The agency said it planned to meet with driverless-system developers before the end of July.

For Zoox, the recall lands during a push to put its purpose-built vehicle in front of more riders. Amazon acquired Zoox in 2020, keeping Levinson and CEO Aicha Evans in charge. Zoox operates as an Amazon subsidiary while pursuing a vertically integrated ride-hailing service built around a four-seat vehicle without a steering wheel or pedals.

Zoox and Uber also announced a distribution partnership in March that calls for Zoox robotaxis to appear in Uber's Las Vegas service, followed by Los Angeles in 2027. The agreement gives Zoox access to Uber's riders while leaving Zoox responsible for the vehicles and autonomous-driving system.

That expansion raises the cost of unusual behavior. A small, tightly supervised fleet can depend heavily on operational controls and human intervention. A larger service needs the driving system to resolve emergency scenes consistently, including scenes that lack clean lane closures or familiar visual cues.

A recurring test for Zoox's software model

This is not Zoox's first software safety action. A December 2025 recall covered 332 vehicles after Zoox identified 62 instances in which its vehicles unnecessarily crossed lane lines at or near intersections. Zoox attributed those events to several causes, including mislabeled double-parked vehicles, unexpected routing changes and poorly executed attempts to avoid blocking cross traffic.

NHTSA also opened a preliminary investigation in May 2024 after two Toyota Highlanders running the Zoox automated-driving system braked suddenly and were struck from behind by motorcyclists. Both crashes resulted in minor injuries, according to the agency.

Those cases involve different driving behaviors, but they illustrate the operating reality Levinson chose when Zoox began building an end-to-end robotaxi. Zoox can deploy software remedies across vehicles it controls and collect fleet data without waiting on consumers. Each intervention also becomes a public record of where the system's judgment fell short.

Levinson's technical background helped Zoox pursue one of the industry's most ambitious product designs: a bidirectional vehicle created around autonomy from its first sketch. The smoke recall shows where that ambition is being measured now. Building a vehicle without a driver removes the human fallback inside the cabin. Zoox must encode the situational judgment that lets an ordinary driver recognize danger before cones, signs or a cleanly labeled route make the answer obvious.