NHTSA Just Built a Safety Test for Cars That Don't Need It

NHTSA just added four real tests to its five-star program. Pedestrian AEB: a forward radar-and-camera system detects a person in the road and applies full braking force in roughly 300 milliseconds. Lane-keeping assistance: torque-vectoring corrections when the car drifts across lane markings. Blind-spot warning and intervention: short-range radar tracks adjacent vehicles and physically resists a lane change into an occupied space.^[1] These are genuine engineering achievements. I've wanted NHTSA to test them for years.

The 2026 Tesla Model Y was the first vehicle to pass all four. It already had the lowest fatality rate of any SUV in FARS: 0.03 deaths per 100 million vehicle miles traveled.^[2] Certifying the Model Y as safe is like giving a structural integrity award to a building that has never had a crack.

Run the FARS model-year death distributions against the ADAS adoption timeline and the gap is staggering. Of the 187,058 vehicle-occupant fatalities in the 2014-2023 dataset, 92.4% involved model years before 2018, the era before automatic emergency braking became standard equipment. Just 7.6% involved vehicles from 2018 or newer. Narrow to model years 2022 and later, where AEB is nearly universal, and you get 0.6% of deaths. The new test regime certifies 0.6% of the problem.

Look at the ten deadliest vehicles by fatality rate. Seven are discontinued: the Hyundai Veloster (8.54 deaths per 100M VMT, killed off in 2021), the Chevy Tracker (7.83, gone since 2004), the Cobalt (5.10, axed in 2010), the S-10 pickup (4.83, dead since 2004), the Cadillac Seville (3.89, discontinued 2004). No airbag recalibration, no sensor fusion retrofit, no over-the-air update will reach these vehicles. They are mechanically frozen in whatever decade they rolled off the line. Meanwhile, the five safest vehicles are current-production, ADAS-equipped models that passed this test in practice years before NHTSA wrote it on paper.

NHTSA finally codified what engineers have known: the technologies that prevent crashes are more important than the structures that survive them. Pedestrian AEB alone, when deployed fleet-wide, could prevent thousands of fatalities annually. That is the right bet. But "fleet-wide" is doing enormous load-bearing work. The average vehicle on U.S. roads is 12.6 years old and climbing, per S&P Global Mobility.^[3] Fleet turnover is decelerating. Every year a 2006 Impala stays registered is a year these tests cannot touch its occupants.

The uncomfortable math: for every fatality in a 2022-or-newer vehicle, there are 143 fatalities in vehicles built before ADAS existed. No test, no rating, no five-star sticker fixes a 2007 Cobalt on a rural two-lane at 11 p.m. with no forward-collision warning and no electronic stability control on the base trim.

Strongest counterargument

NHTSA tests new vehicles because that is what regulators can control. You cannot mandate a radar module onto a 2004 S-10. The point is not to fix the current fleet but to shape the future one: by demanding standard ADAS on new cars, regulators accelerate the day when the dangerous tail of the fleet finally ages out. This is correct, and the new tests will save lives over a 15-to-20-year horizon. But the 143:1 ratio means the deaths are happening now, and every policy conversation that celebrates the benchmark without acknowledging the fleet-age gap risks confusing progress on paper with progress on pavement.

What this means for you

If your vehicle predates 2018, you are in the fleet where 92% of fatal crashes occur. NHTSA's new tests will not help you. What will: check your VIN at nhtsa.gov/recalls for unfixed safety recalls. Aftermarket blind-spot mirrors cost $8. A forward-facing dashcam with collision warning runs under $100. These are not ADAS, but they are the interventions available to the overwhelming majority of the fleet that NHTSA's testing program cannot reach.

Limitations

The 92.4% figure is cumulative across the 2014-2023 FARS window. Older model years have more exposure time in the dataset (a 2005 model accumulates crashes over all ten years; a 2022 model has one or two). This inflates the raw count for older vehicles. It does not change the core finding: the fleet currently killing people is overwhelmingly pre-ADAS, and no new-car test program addresses that retroactively. A rate-controlled analysis would narrow the gap but not close it.