Alphabet's Waymo filed a voluntary recall covering nearly 3,800 autonomous vehicles after one of its driverless cars drove straight into a flooded San Antonio road on April 20 and was swept away, a software failure that raises hard questions about whether regulators and the public are getting honest answers about what these machines can and cannot do.
No passengers were inside the vehicle when it entered the floodwater. No one was injured. But the fact that a car built to navigate public streets on its own could not recognize an impassable road and stop is not a minor glitch. It is exactly the kind of failure that shakes whatever public confidence remains in the rush to put autonomous vehicles on American roads.
The recall notice, first reported by the Daily Mail, appeared on the National Highway Traffic Safety Administration's website and warned of "a loss of vehicle control, increasing the risk of a crash or injury." The recall covers vehicles running Waymo's fifth- and sixth-generation automated driving systems.
On April 20, an unoccupied Waymo robotaxi approached a flooded roadway in San Antonio. The Washington Times reported that the vehicle slowed but did not stop, and was swept into a creek. The company suspended San Antonio service afterward, and that suspension remains in place.
Waymo says it initiated the recall on April 24, just four days after the incident. The company told the Daily Mail it had "identified an area of improvement regarding untraversable flooded lanes specific to higher-speed roadways" and chose to file the voluntary recall with NHTSA.
A spokesperson added that the company is "working to implement additional software safeguards and have put mitigations in place, including refining our extreme weather operations during periods of intense rain, limiting access to areas where flash flooding might occur."
That language, "area of improvement", is the kind of corporate euphemism that should make regulators and riders pay closer attention. A car that drives into a creek is not an "area of improvement." It is a failure of the core promise: that these vehicles can safely operate without a human behind the wheel.
Newsmax reported that the recall covers about 3,800 vehicles, all using the fifth- and sixth-generation systems. The NHTSA recall documents confirm the affected vehicles were built from 2022 through April 20 of this year. Waymo says a software fix is under development, and it plans to resume San Antonio service once the update is deployed.
Waymo currently provides more than 500,000 trips per week across its operating cities, which include San Francisco, Austin, and Miami. That volume makes the stakes clear. A software bug that sends a car into standing water in one city could, in theory, do the same in any city where these vehicles operate.
The autonomous vehicle industry has faced a steady stream of recalls tied to software-related safety problems across multiple companies. The pattern is familiar: a failure surfaces, the company frames it as voluntary and proactive, regulators post the notice, and the public is left to decide whether the technology is ready for the roads they share.
The San Antonio flood incident is not an isolated episode. In December, a blackout in San Francisco caused Waymo taxis to stop in place, creating significant traffic disruption. In central Phoenix, a Waymo vehicle was trapped by floodwater last year. And the problems are not confined to the United States.
Last month, a Waymo reportedly crashed through a police cordon in Harlesden, northwest London, where officers were responding to a double stabbing. In another London neighborhood, Spitalfields, residents of one cul-de-sac say they are repeatedly woken at 4 a.m. on most weekdays by Waymo taxis getting trapped in their street.
These are not catastrophic crashes. But taken together, they paint a picture of a technology that still struggles with the messy, unpredictable conditions of real-world driving, exactly the conditions a machine must handle if it is going to replace a human driver.
The broader autonomous vehicle sector continues to expand rapidly. Competitors like Zoox are adding new cities to their testing footprints, and the regulatory framework has not kept pace with deployment.
Professor Jack Stilgoe, an expert on emerging technologies at University College London, told the Daily Mail that the response to incidents like this one matters as much as the incidents themselves:
"Companies need to show that they aren't reckless. How they respond to mishaps and crises is absolutely vital."
Stilgoe also acknowledged the inherent limits of the technology. "No technology will ever be perfectly safe," he said. "Self-driving vehicles have to operate in unpredictable public spaces with other road users, surrounded by all of the complexity of everyday life."
That is a fair point. But it cuts both ways. If the technology cannot be perfectly safe, then the companies deploying it owe the public far more transparency than corporate jargon about "areas of improvement." And regulators owe the public more than a website posting.
Professor Siddartha Khastgir, head of Safe Autonomy at the University of Warwick, put it more directly:
"Public trust is underpinned by transparent, honest and responsible communications. Like any technology, the concept of absolute safety is a myth. But we can still have safe autonomous vehicles if we can accurately establish and communicate their capabilities and limitations to the users, we call this 'informed safety.'"
"Informed safety" is a useful concept, and one that the autonomous vehicle industry has not embraced with anything approaching the rigor it demands. Consumers are told these vehicles are safe. They are not told, in plain language, what happens when the software encounters a flooded road, a police cordon, or a power outage.
The accountability gap extends well beyond robotaxis. Tesla owners have filed class-action lawsuits alleging the automaker misled buyers on self-driving capabilities, and the broader question of who bears responsibility when autonomous systems fail remains largely unresolved in the United States.
Waymo intends to launch a robotaxi service in London starting in September, following a trial period. Whether this recall will affect those plans remains unclear. In the United Kingdom, legal responsibility after an autonomous vehicle crash falls under the 2024 Automated Vehicles Act, a framework that at least attempts to address the "grey area" Stilgoe described.
"Even though we've seen lots of these things moving around US and Chinese cities, responsibility is still a grey area," Stilgoe said. That grey area is a problem for every city where these vehicles operate, and it will only grow as fleets expand.
Stilgoe added that regulators have a role to play: "Things will go wrong. But regulators can help minimise the risks and put in place processes so that, when mistakes happen, everyone can learn from them."
The question is whether American regulators are doing that. NHTSA hosted the recall notice on its website. But a notice on a government PDF is not the same as meaningful oversight of a rapidly scaling technology that shares roads with millions of human drivers, pedestrians, and cyclists.
Vehicle recalls have become routine across the industry, from mechanical defects in new vehicle platforms to software problems affecting hundreds of thousands of cars. What distinguishes the Waymo recall is the nature of the failure: a vehicle with no human driver made a decision that any human driver in Texas would know not to make.
Waymo frames this recall as evidence of responsible self-regulation. The company identified the problem, filed the notice, suspended service, and says it is working on a fix. That is better than silence.
But the underlying issue is not whether Waymo responded appropriately after the fact. The issue is that a vehicle operating on public roads, carrying the implicit promise that it can drive itself safely, could not distinguish a passable road from a flooded creek. And this is not the first time a Waymo vehicle has encountered floodwater and failed to respond correctly.
The New York Post noted additional flood-related problems in Texas and Los Angeles, suggesting the software gap was not limited to a single incident or a single city.
Nearly 3,800 vehicles. More than 500,000 trips per week. A London launch planned for September. And a software system that, until this recall, could send a car into a flooded road at highway speed.
The technology may get there eventually. But the public deserves more than corporate euphemisms and PDF filings when a driverless car ends up in a creek. If "informed safety" means anything, it starts with admitting what the machine cannot do, before it proves the point on its own.