Waymo's robotaxi is safer than you, until it hits a work zone

In Phoenix, a blind man opens the door, settles into the back seat, and no one is behind the wheel. A voice reads out the route, a chime guided him to the car, the seat leaves room for his guide dog. He sets off alone, with no driver to negotiate with, no favour to ask of a relative. For someone who has never been able to drive, an ordinary trip becomes possible.

On 18 June 2026, that same week, the US federal road-safety agency published a recall covering 3,871 Waymo robotaxis. The reason: on the freeway, the fifth-generation software could enter a construction zone at full speed, failing to recognise it in time. Both facts describe the same technology, and it is precisely their coexistence that is worth pausing on.

Driving better than we do, by the numbers

Waymo's safety record is not a promise, it is a count. The fleet has passed 100 million driverless miles, at close to four million miles a week in early 2026. Across 170 million strictly autonomous miles, the company reports 0.02 serious-injury crashes per million miles, against 0.22 for human drivers in the same cities. That is 92% fewer serious crashes.

The figures hold for the most vulnerable too: 92% fewer pedestrian injuries, 85% fewer for cyclists. A machine that does not get distracted, does not drink, does not run a light to save thirty seconds, mechanically removes a share of road violence. On this ground the robotaxi does not merely match the human, it does markedly better.

What those numbers cover still matters. Most of the miles were driven on urban streets, in daylight, in clear weather, and compared with local benchmarks rather than a national average. The performance is real, but it is measured where driving is easiest. That is a qualification, not a rebuttal, and it sets up what follows.

For those who cannot drive, mobility returned

This is probably where the technology changes lives most, and least visibly. Blind and low-vision riders describe the robotaxi as a clear shift in daily life: leaving when they want, without depending on a relative, without enduring a hurried or reluctant driver. Working with schools for the blind in Texas from the start rather than as an afterthought, Waymo built exterior audio cues to locate the car, voice narration of the trip, and an option that stops the vehicle on the rider's side of the street to cut down on walking.

The benefit is measured in time and in mental load. An older person who no longer drives, a night-shift worker, a parent without a second car regain a freedom of movement once reserved for those holding a wheel. The National Federation of the Blind, long cautious, now speaks of a turning point. That is no small thing: mobility shapes employment, healthcare, social life.

This gain has a less flattering name on closer inspection: delegation. You hand your journey to a system you neither steer nor understand. As long as it works, it is comfort. The day it stops, you measure how much you had leaned on it.

The recall, or the conditions of the promise

That is exactly what the spring of 2026 showed. Waymo's freeway service, opened only in November 2025, was suspended on 19 May in San Francisco, Los Angeles, Phoenix and Miami. A month later the recall made the cause official: faced with several hazards at once, the software could rank its priorities wrongly and cross a work zone at speed.

The episode is instructive less for its severity than for its nature. The car did not err at random; it stumbled on an unusual, poorly learned situation, because freeways and their moving work zones form an environment rarer and more treacherous than the street. The safety of an autonomous system is never general: it holds for the cases it has seen a thousand times, and wavers before those it has rarely met.

The recall also bears concretely on the promise of freedom. The freeway was the piece Waymo counted on to move from neighbourhood to region, and to target more than twenty new cities in 2026, from Denver to London. Until the fix is validated, the robotaxi stays a downtown tool, remarkable but confined.

Depending on an operator, and on its map

Finally, it is worth looking at whom these trips are entrusted to. A robotaxi is not a car you own: it is a service, tied to an app, a company, an area it has chosen to cover. Outside the handful of served cities, the promised freedom simply does not exist. And where it does, it comes with sensors filming the street continuously and a journey on record.

The autonomy returned to the passenger thus comes paired with dependence on the operator. You gain independence from the wheel and lose a little to the provider, its pricing, its outages and its data policy. It is not an absurd bargain, especially for those who never had the option of driving, but it is a bargain best made with eyes open.

The robotaxi keeps part of its promise, and keeps it seriously: it drives more carefully than we do and opens the city to those it kept at a distance. The spring recall is the other half of the sentence. This freedom is conditional, bounded by a map, by the weather, by the situations the machine has learned to handle. The question is not whether we want a flawless chauffeur, there is none, but how far we accept that our mobility hangs on a system whose limits of competence we cannot, in truth, see.