Waymo, Alphabet’s autonomous driving subsidiary, has recalled nearly 3,800 robotaxis after one of its self-driving vehicles entered floodwater during severe storms in San Antonio, Texas, reviving a debate that has followed the autonomous vehicle industry for years. How safe do self-driving cars actually need to be before the public is willing to trust them?
According to filings submitted to the National Highway Traffic Safety Administration, the company identified a software issue that could allow Waymo vehicles to “slow but not stop” after detecting potentially impassable flooded roadways on higher-speed streets. One vehicle was ultimately swept into Salado Creek during April flooding in San Antonio after proceeding into standing water. No passengers were inside the vehicle, and no injuries were reported.
Waymo described the issue as “an area of improvement regarding untraversable flooded lanes specific to higher-speed roadways,” adding that the company had implemented “additional software safeguards” and refined its severe weather operations.
The recall itself is technically straightforward. Because Waymo’s fleet operates through software-defined systems, the company was able to distribute over-the-air updates across thousands of vehicles without requiring dealership repairs or physical recalls in the traditional automotive sense. The broader implications are far more complicated.
Autonomous driving technology has now reached a point where the central question is no longer whether self-driving cars can function on public roads. In cities including Phoenix, San Francisco, Los Angeles, and Austin, Waymo vehicles already complete hundreds of thousands of rides every week with no human driver behind the wheel. The technology works well enough to operate commercially at scale.
What remains unresolved is whether society is prepared to judge machine-driven transportation against the same standards long applied to human drivers.
Also Read: Will Walmart’s Digital Price Tags Herald the Death of Fixed Prices?
Human Drivers Already Produce Catastrophic Outcomes
More than 40,000 people die annually on American roads, according to federal transportation data, with the overwhelming majority of crashes linked to human error, including distraction, intoxication, speeding, fatigue, and reckless driving. Autonomous vehicle companies frequently frame their technology around that reality. The argument is not that self-driving systems will become perfect. The argument is that they may eventually become significantly safer than ordinary human behavior behind the wheel. Statistically, that may already be partially true.
Waymo-backed safety research published earlier this year found substantially lower injury-related crash rates compared to human drivers across millions of autonomous miles traveled. At the same time, the company continues facing recalls, federal scrutiny, and highly publicized operational failures that shape public perception far more powerfully than broad safety statistics.
That disconnect sits at the center of the autonomous vehicle debate.
A fatal crash caused by a drunk driver rarely becomes a national referendum on whether human beings should continue operating vehicles. Human driving is deeply normalized, even though it produces enormous casualty figures every year. Autonomous vehicle incidents are treated differently because machine error feels fundamentally less acceptable to the public, particularly when the technology is marketed as safer and more advanced than human judgment.
The floodwater recall illustrates this tension clearly. One empty robotaxi entering standing water generated national headlines and federal attention, while thousands of fatal crashes caused by human drivers this year will pass with comparatively limited scrutiny.
The Problem Is Not Intelligence Alone
Flooded roads expose one of the hardest problems in autonomous driving because severe weather disrupts many of the visual and environmental assumptions these systems rely upon.
Standing water can obscure lane markings, alter reflections, conceal roadway damage, and interfere with the interpretation of lidar, radar, and camera systems. Human drivers often rely on contextual reasoning during storms by observing stalled vehicles, water movement, surrounding traffic behavior, debris patterns, and local road familiarity. Autonomous systems process those same environments through probabilistic software models trained on enormous amounts of driving data. That distinction becomes critical when roads stop behaving predictably.
Igal Raichelgauz, CEO of autonomous driving company Autobrains, recently argued that current systems still lack “common sense” during unusual roadway situations. The comment followed the San Antonio flooding incident and reflects a growing concern within the industry itself that machine learning systems still struggle when environmental conditions fall outside their expected operational patterns.
The autonomous vehicle industry has spent years refining systems capable of navigating normal traffic conditions with remarkable consistency. Edge cases remain the real obstacle. Flooding, construction zones, emergency scenes, damaged infrastructure, severe storms, police detours, and unpredictable human behavior continue exposing weaknesses that engineers must solve incrementally through software updates and operational restrictions.
Waymo’s recall filing itself acknowledged this limitation directly, stating that vehicles may continue into “potentially untraversable flooded lanes” under certain conditions.
The Industry Is Being Asked to Achieve Near-Perfection
The autonomous vehicle industry increasingly finds itself operating under a standard that traditional transportation systems have never faced.
Human-driven transportation already produces mass casualties on a scale society has largely normalized. Yet autonomous systems are often treated as unacceptable if they produce even isolated high-profile failures. That expectation may be understandable given the stakes involved, but it also creates a threshold that no transportation technology has ever realistically achieved.
None of this eliminates the need for aggressive oversight. Autonomous vehicle companies are deploying experimental technologies onto public infrastructure, often in densely populated cities. Federal investigations into Waymo and other self-driving companies have continued expanding in recent years as regulators examine collisions, roadway behavior, school zone incidents, and emergency response interactions.
At the same time, the industry’s critics and supporters are increasingly arguing over two fundamentally different visions of transportation risk. One side sees autonomous driving failures as evidence that the technology remains too dangerous for public deployment. The other sees human driving as an already catastrophic system that society tolerates largely because its dangers feel familiar.
The uncomfortable reality underlying the autonomous vehicle industry is that self-driving systems may eventually become statistically safer than human drivers long before the public becomes emotionally willing to accept machine-caused deaths at all.
