NHTSA Stamped ‘100% Defect Rate’ on Waymo. It Accidentally Proved Software Recalls Are Better.

On April 20, an empty Waymo robotaxi in San Antonio detected a flooded road ahead, slowed from highway speed, and drove into the water anyway, where the creek swept it downstream like a bathtub toy on the San Antonio River. Nobody was aboard, nobody was hurt, and Waymo pushed an over-the-air software fix to every vehicle in its fleet that same day. Three weeks later, NHTSA filed the recall paperwork. The official designation: 3,791 vehicles affected, 100% defect rate.

That number was engineered to sound catastrophic, and it worked. Headlines screamed about Waymo's entire fleet being defective. Fox News quoted unnamed sources saying the company was "putting American lives at risk." What none of those headlines mentioned: a 100% defect rate is the default for any software bug. If one car runs the buggy code, every car running that code has the same bug. That is not a scandal; that is how computers work.

Compare it to what happens when the recall system encounters hardware, which is the thing it was actually built for. In 2025, manufacturers issued approximately 1,000 separate recalls covering 31.3 million vehicles across the United States, according to NHTSA data compiled by autoinsurance.com.^[1] Ford alone accounted for 153 recalls touching 12.9 million vehicles. The average recall completion rate across all manufacturers that year was 45%.^[1] Not 45% within the first month; 45% ever. More than half of recalled vehicles never get fixed, and NHTSA knows this because they track it and publish the numbers in their annual reports.^[2]

The gap widens with vehicle age. NHTSA's own data shows 76% completion for vehicles one to five years old, dropping to 56% for six-to-ten-year-old vehicles.^[2] Beyond ten years, the agency stops publishing the number, which tells you what it looks like. The Takata airbag recall, which ultimately affected tens of millions of vehicles, took over a decade and the completion rate for older models plateaued below 50%.^[3] People drove cars with literal fragmentation grenades behind their steering wheels for years because the recall system runs on postal mail and dealer appointments.

Waymo's fleet was fixed before the recall letter was drafted, every vehicle updated on the same day without a single dealer appointment. NHTSA sent its formal request on May 11, three weeks after the fix was already deployed.^[4] The regulatory paperwork outlived the safety risk by twenty-one days.

Now run the effective risk math on a hypothetical fleet of 100,000 vehicles with an identified safety defect. Under the traditional hardware recall pipeline, approximately 45,000 get fixed over a period of months to years; 55,000 remain on public roads with a known defect indefinitely. Under the software OTA model, approximately 100,000 get fixed within hours to days; zero remain defective. The "100% defect rate" recall eliminates more risk, faster, with more certainty than any hardware recall in NHTSA's history. Sidley Austin, in their legal analysis of the NHTSA recall order, noted that in 2025 the agency processed 997 recalls and 88% were manufacturer-initiated without any NHTSA involvement at all.^[5] The system runs on voluntary compliance and incomplete execution.

The strongest case against this framing: software failures are correlated in a way hardware failures are not. When a batch of brake pads has a metallurgical defect, each pad fails independently under different conditions at different times. A software bug is deterministic: if the conditions are met, every vehicle fails identically, simultaneously. The San Antonio flood was benign because the car was empty and the failure mode was "drove into water." A different bug, under different conditions, could trigger simultaneous failures across an entire fleet. That correlated risk profile is genuinely new, and the recall system's instinct to flag it is not entirely wrong.

But correlated risk is a deployment architecture problem, not a recall system problem. You address it with staged rollouts, fleet-wide telemetry, geofencing, and operational design domains, which Waymo already uses. You do not address it by stamping "100% defect rate" on a filing and calling it a day while 17 million vehicles from a single manufacturer sit unfixed in driveways because their owners threw the recall notice in the trash.

In 2022, NHTSA tracked 23 over-the-air software recalls, double the prior year's count.^[6] As vehicles become increasingly software-defined, the share of recalls fixable via OTA will grow. The question is whether NHTSA will update its vocabulary to reflect the actual risk calculus, or whether "100% defect rate" will continue to be deployed as rhetorical ammunition against the vehicles most capable of fixing themselves.

If you own a car right now, go to nhtsa.gov/recalls and enter your VIN. There is roughly a one-in-three chance you have an open recall you never completed, and no one is coming to remind you again. When you shop for your next vehicle, ask whether it supports over-the-air software updates; the answer tells you whether future safety fixes require your participation or happen while you sleep. And the next time a headline screams about a "100% defect rate" on a software-defined vehicle, read the completion rate column. A recall that fixes every car in hours is not a scandal. A recall system that leaves seventeen million cars broken on the road permanently is.