Waymo "Fixed" Its Software, and the Robotaxis Kept Going Around Stopped School Buses: the Regulator Says Most Incidents Came After the Fix

Almost a year ago Waymo issued a recall and updated its self-driving software after reports piled up that the company's robotaxis were illegally passing school buses stopped with their red lights flashing. Now the US National Highway Traffic Safety Administration (NHTSA) is opening a new investigation, because the behaviour didn't stop.

In a filing dated 6 November, NHTSA states that most of the reported incidents happened after the update that was supposed to solve the problem. Waymo had claimed the fix was completed on all affected vehicles on 17 November 2025. The Office of Defects Investigation (ODI) has now opened a so-called recall query to "assess the adequacy of the remedy".

The first investigation began in October 2025, after a video from Atlanta. The bus was stopped with its stop arm out and lights on, and children were getting off. The robotaxi crossed in front of the bus from its right side, then turned left around the front and drove on down the street. More cases followed in Austin, where Waymo also carries passengers. The local school district reported at least five such incidents in a letter to NHTSA.

The company responds with standard PR lines. "We have made meaningful improvements to how our vehicles behave around school buses and have had no citations since March 2026," a spokesperson said. "The Waymo Driver has a clearly demonstrated positive impact on road safety." Waymo says it will respond to the query and looks forward to further cooperation on recalls "for autonomous vehicle software that is constantly learning and improving".

That last sentence is the key. With an ordinary car, a recall means a replaced part and a solved problem. With a robotaxi, the "fix" is a new version of software that learns, so it is hard to claim in advance that the lesson has been learned. A school bus is the clearest signal on the road: red lights, an extended sign, children. If the algorithm stumbles exactly there, what's left for more complicated situations? And who answers when nobody is behind the wheel?