Zoox, a self-driving taxi company owned by Amazon, recalled its entire fleet of 105 vehicles on July 17, 2026, after one of them drove into heavy smoke at an active fire scene and could not get out on its own. The recall, registered under NHTSA number 26E044000, follows a June 20 incident in which a Zoox taxi encountered thick smoke from a fire burning in a traffic lane that had not been blocked off with cones (Engadget).

The vehicle drove into the smoke, braked hard while trying to steer away, and came to a stop. A remote operator — a human who can step in and help Zoox vehicles in difficult situations — guided the taxi to back out of the area. After that, firefighters placed traffic cones blocking two of three lanes, setting up a boundary the vehicle had not noticed by itself (Engadget).

Zoox said this was the only time one of its vehicles had a smoke-related problem like this. The software update released with the recall "enhances existing capability of detecting and responding to heavy smoke," according to the company. The NHTSA recall report notes that the taxi's interference with the fire emergency is what prompted the update and recall (NHTSA).

This is Zoox's second software recall. In May 2025, the company recalled vehicles after one of its Las Vegas taxis collided with a passenger car (Engadget).

The recall comes as regulators are paying closer attention to how self-driving vehicles behave around emergency scenes. In early July, NHTSA — the U.S. agency that oversees vehicle safety — published a letter demanding that autonomous vehicle companies address the problem of self-driving taxis interfering with emergency response operations. The agency said it plans to meet with AV companies before the end of July to discuss the issue (Engadget).

NHTSA Administrator Jonathan Morrison was blunt in that letter. "The inability to detect and appropriately respond to such situations represents a functional insufficiency," he wrote, adding that "emergency scenes are not rare or extreme edge cases" (Engadget).

The regulatory pressure coincides with Zoox's expansion plans. In March, the company announced it would expand its service area in Las Vegas and San Francisco and begin testing in new cities (Engadget).

There is a deeper shift happening in how regulators expect the self-driving car industry to think about emergency scenes. The industry has long treated situations like fires and accidents as edge cases — meaning rare, unusual scenarios that fall outside what a self-driving system normally handles and can be fixed gradually over time through software updates and more driving data. NHTSA's letter explicitly rejects that view, saying emergency scenes should be treated as a basic requirement, not a future improvement.

The June 20 incident shows why this matters. The Zoox vehicle did not detect the smoke as something it should stop for or avoid before driving into it. The car braked and swerved only after it was already inside the smoke, which means its sensors and software failed to recognize the danger soon enough to respond calmly. The fact that a remote human operator had to step in and reverse the vehicle confirms the car could not solve the problem on its own.

Smoke is a uniquely difficult challenge for self-driving cars. Most of the sensors these vehicles rely on — like lidar, which uses laser beams to detect objects, and cameras — are designed to spot solid things like other cars, people, or barriers. Smoke is made up of tiny floating particles. When a laser beam or camera hits dense smoke, it produces fuzzy, unclear readings rather than the sharp signal you get from a solid object. The car's software cannot easily decide whether the smoke is something to avoid or just empty space, which is the kind of decision these systems need to make to plan a safe route.

One thing worth noting is that Zoox fixed this with a software update, not by adding new hardware. That suggests the company improved how the car's existing sensors interpret what they see, rather than giving the vehicle new types of sensors. This matches Zoox's description of the fix, but it also means the solution is limited by what the current sensors can actually detect in heavy smoke.

Recalling an entire 105-vehicle fleet over a single incident is also telling. With a fleet that small, one failure works out to roughly a one-percent rate for this type of scenario. Zoox's decision to recall all vehicles suggests the company found a broad gap in how its software handles smoke, rather than a one-off glitch. NHTSA's demand that the whole industry pay attention to emergency-scene interference signals that regulators see this as a problem affecting the entire category, not just Zoox.

For the self-driving car industry, the direction is clear. Companies are expanding into busier cities where emergency scenes — car fires, building fires, accident responses — happen regularly. A system that treats these as rare edge cases will run into them more and more as the number of vehicles grows. NHTSA's position, as Morrison put it, is that handling emergency scenes is a prerequisite for operating at scale, not something to improve later. Zoox's recall is the first concrete example of that principle being enforced, but the agency's planned meetings with AV companies before the end of July suggest it will not be the last.