Waymo Is Telling Regulators It Cannot Yet Fix Its Flooding Problem
Waymo Is Telling Regulators It Cannot Yet Fix Its Flooding Problem
Waymo has paused robotaxi service on freeways in four cities and pulled out of two others. The reason is a familiar one: its vehicles keep driving into flooded roads. But the more telling detail is buried in a regulatory filing. The company's actual software fix for the problem is, as of mid-May, still under development.
The NHTSA recall filing from May 11, 2026 covers 3,791 fifth- and sixth-generation robotaxis. It describes the defect plainly: the software may allow a vehicle to slow and then drive into standing water on higher-speed roads. The filing's remedy section, however, consists of a single sentence that tells you everything about Waymo's current position: "The remedy is currently under development." As an interim measure, Waymo has modified the scope of vehicle operations to add weather-related constraints. In other words, geo-fencing. The cars are being told where not to go, rather than being taught to recognize water.
The contrast between Waymo's public posture and its technical reality is the story here. The company is simultaneously operating thousands of vehicles with a known software defect it has not yet patched, pursuing a goal of one million paid rides per week by the end of 2026, and expanding aggressively in Houston. The recall does not describe a hypothetical risk. It documents incidents that have already happened.
In San Antonio on April 20, an unoccupied Waymo vehicle entered a flooded road and was swept into Salado Creek. The car was recovered. In Atlanta on May 20, another unoccupied Waymo vehicle got stuck in a flooded road during a storm and sat there for roughly an hour before being retrieved. On January 26 in Austin, a Waymo vehicle struck a child near a school, braking from approximately 17 mph to under 6 mph before impact; the child sustained minor injuries. NHTSA is investigating that incident. NTSB is separately investigating reports of Waymo robotaxis illegally passing unloading school buses in Austin. The school bus and child incidents predate the flooding cluster, which suggests this is not a single-parameter failure.
Waymo is not small. The company says it now provides more than 500,000 trips per week across multiple cities, a volume it has built over several years of careful geographic expansion. It started offering highway rides in late 2025 and is targeting World Cup transportation readiness. It is expanding Houston service to nearly 50 square miles including the Medical Center and NRG Park. These are not startup ambitions. They are infrastructure-scale plans backed by Alphabet.
The freeway suspension in San Francisco, Los Angeles, Phoenix, and Miami affects one of Waymo's newer service categories. The company has not disclosed how many rides per week occur on freeways versus surface streets, so the direct rider impact of this suspension is unclear. What is clear is that the geographic expansion is running ahead of the technical remediation. The fix is not ready. The fleet is growing anyway.
There is also a passenger video that does not fit neatly into the recall narrative. On May 19, a Waymo rider in Los Angeles posted footage of his vehicle blasting through traffic cones on a freeway, swerving around large trucks, and accelerating away from a police stop. "These are not ready for highways," he wrote. Waymo has not publicly addressed that specific incident in the context of the freeway suspension.
The NHTSA investigation into the Austin child incident is ongoing. The NTSB school bus investigation is ongoing. The recall remedy is under development. These three open processes mean the full regulatory picture of Waymo's current safety posture will not be clear for some time. In the meantime, Waymo's expansion continues in Houston and elsewhere, and the question of whether the company can actually teach its cars to understand water remains unanswered.
The person next to the robotaxi is, as always, part of this story. So is the taxpayer, the city planner, and the insurance underwriter. Waymo's confidence in its system is not irrational. The incident rate per mile is, by the company's own reporting, lower than the human baseline. But confidence and competence are different things when the system has a known defect that has not been fixed. Geo-fencing is a reasonable interim response. It is not a solution.