Waymo is the leading self-driving taxi company in the world, owned by Alphabet (Google's parent), and currently the only Western company running paid, fully driverless rides at meaningful scale. As of this spring, it delivers roughly 500,000 paid rides a week across 10 US cities. In February, it raised $16 billion at a $126 billion valuation. By any measure, it is the front-runner in a race that many traditional transport, logistics, and insurance businesses are watching closely.
So when two cities go dark in the same month because the cars keep driving into floods, it matters.
The first incident happened in San Antonio on April 20. An unoccupied Waymo car drove into a flooded section of a 40-mph road, detected the danger, but slowed down instead of stopping. The car was swept into Salado Creek and had to be recovered four days later from a park downstream. Waymo paused all San Antonio operations, which turned into its longest-ever stoppage in that city. On April 30, it filed a formal recall with US road safety regulators covering the entire US fleet of 3,791 vehicles. The fix was pushed to the cars remotely, the way a phone gets a software update, so no cars had to go to a workshop.
But the company admitted in the recall filing that the fix was not a finished solution. It placed restrictions on areas and times with a higher chance of flooding, and said engineers were still working on the full repair. That interim fix failed in Atlanta this week, when another unoccupied car entered a flooded intersection during a rainstorm so sudden it preceded any weather warning from national forecasting services. The car was stuck for about an hour before being recovered.
The core problem here is that these cars rely on signals, sensors, and pre-mapped data to read the road. They are very good at many things humans are bad at: staying in lane, keeping distance, not getting distracted. But flooding is different. It can appear in minutes, on a road that looked completely normal an hour before. The car's system slowed down when it sensed water, which shows it detected the problem, but it did not stop or reroute. That gap between detecting a problem and acting on it correctly is exactly the kind of edge case that takes years to close.
This is not the first time Waymo has struggled to stamp out a problem after issuing a fix. From September 2025 onward, its cars were repeatedly caught passing stopped school buses in Austin and Atlanta, which is illegal. A software recall was issued in December for 3,067 vehicles. Violations continued after the recall. The US road safety board opened a formal investigation in January, as did a second federal regulator. Those investigations are still ongoing.
In one of the school bus incidents, investigators found that Waymo's car had correctly stopped and asked a remote human operator whether the signals were active. The remote operator said no. The car then proceeded past the bus while children were boarding. That detail reveals something important: at edge cases, these systems still depend on human judgment, and that judgment can be wrong.
Waymo's overall safety record, across more than 127 million fully autonomous miles, is significantly better than the average human driver. Serious injury crashes are down roughly 90% compared to human benchmarks. The company is not hiding from scrutiny. It files voluntary recalls, cooperates with regulators, and pushes fixes to its whole fleet simultaneously, which is something no traditional car company can do. When Toyota issues a recall, it takes months for vehicles to actually be fixed. Waymo's fleet is updated within days.
But the flooding and school bus episodes, happening during the company's fastest-ever growth phase, raise a fair question for any business operator thinking about where self-driving transport is in its maturity: the technology works extremely well in normal, predictable conditions, and it still has real gaps when conditions are unusual or change faster than the system expects.
For businesses in transport, logistics, or insurance, the practical read is straightforward. Fully autonomous vehicles are real, commercially operational, and statistically safer than humans across millions of trips. They are not yet able to handle every edge case, and the edge cases, when they happen, are public and heavily scrutinised. The road to full reliability will be made up of incidents like these, each one closing a gap that was not known until the real world exposed it.