Waymo is recalling 3,791 robotaxis following an incident where a vehicle drove onto a flooded roadway.

      Waymo is recalling 3,791 robotaxis in the U.S. following the identification of a software defect by federal regulators that may cause the vehicles to enter flooded roads at higher speeds, according to the National Highway Traffic Safety Administration (NHTSA) on Tuesday.

      The recall includes vehicles equipped with both fifth- and sixth-generation Waymo Driver, the automated driving system developed by Alphabet. The NHTSA stated that Waymo has adjusted its operational limits related to weather conditions and updated its maps as a temporary measure while engineers work on a permanent software solution. A full resolution is still in development.

      This recall stems from an incident on April 20 in San Antonio, where an unoccupied Waymo vehicle encountered an “untraversable flooded section of a roadway,” as noted in the company’s filing with the NHTSA, and ventured into the standing water at a lower speed instead of rerouting. There were no injuries reported. Waymo voluntarily filed the recall ten days later, on April 30.

      A spokesperson for Waymo remarked that the company had found a specific area for improvement concerning untraversable flooded lanes on higher-speed roads, which led to the recall. The company also mentioned that it was putting in place additional safeguards, enhancing its operations during severe weather, and restricting access to regions prone to flash flooding.

      While the recall may seem small in absolute terms, the filing reveals significant information. By listing every vehicle affected, Waymo has effectively disclosed the size of its U.S. fleet for the first time. The 3,791 figure encompasses deployments in cities like Phoenix, San Francisco, Los Angeles, Austin, San Antonio, and Atlanta. The company had previously confirmed surpassing the 2,000-vehicle mark in September 2025, indicating that the fleet has expanded by nearly 90% in about eight months.

      This positions Waymo as the largest commercial autonomous ride-hailing operation globally. However, it also means that any edge-case failures now carry greater reputational and regulatory implications. This flood recall marks the second software recall Waymo has initiated in six months. In December, the company recalled over 3,000 vehicles to resolve a distinct software problem that caused robotaxis to bypass stopped school buses with their warning lights active.

      Waymo is also under a preliminary NHTSA investigation related to a January event where one of its driverless vehicles collided with a child near Grant Elementary School in Santa Monica during morning drop-off, resulting in minor injuries to the child. The NHTSA's Office of Defects Investigation is assessing whether the system acted with adequate caution considering its proximity to the school, the presence of a crossing guard, and double-parked vehicles nearby.

      Additionally, the National Transportation Safety Board has opened a concurrent inquiry into reports of Waymo vehicles passing stopped school buses in Austin and Atlanta.

      The 3,791 figure also highlights a broader regulatory scrutiny of the autonomous vehicle industry. The NHTSA recently began investigating Uber’s AV partner following 16 crashes in four months in Dallas, using language that was more direct than typical for the industry, indicating “excessive assertiveness and insufficient capability.”

      In China, a system malfunction in April resulted in over 100 Baidu Apollo Go robotaxis becoming immobilized in traffic in Wuhan. This situation underscores that autonomous fleets experience failures differently, and the scale of these incidents challenges existing recall frameworks.

      Waymo maintains, supported by its data, that the Waymo Driver is safer per mile than human drivers in the areas where it operates. The recall announced on Tuesday does not contradict this claim. However, by providing a precise count of the fleet, it raises pertinent questions about how regulators should react as commercial autonomous operations reach a point where edge cases are no longer merely statistical oddities.