America’s Cars Have Never Been Older. NHTSA Says We’ve Never Been Safer. Both Can’t Be True.

NHTSA published its 2025 traffic fatality estimates in April and declared the result “the second-lowest fatality rate in recorded history.”^[1] That would be 1.10 deaths per hundred million vehicle miles traveled, a decimal so polished you could mount it on a plaque. Now hold it up against this one: 36,640 people died on American roads last year, roughly 544 more than in 2019, the last full pre-pandemic year. Rate went down, bodies did not, and somewhere in Washington a press release went out with an exclamation point.

Denominator inflation explains most of the improvement, and the math is embarrassingly simple. Americans drove 29.8 billion more miles in 2025 than in 2024.^[1] Divide a nearly flat death count by a steadily inflating mileage figure and the rate shrinks through pure arithmetic: no lives saved, no engineering breakthrough, just more asphalt under more tires producing a friendlier fraction. It is the statistical equivalent of celebrating your cholesterol ratio improving because you gained sixty pounds, except instead of cholesterol the numerator is corpses.

Meanwhile, the fleet doing most of the dying has never been older. S&P Global Mobility’s 2025 study pegged the average light vehicle on American roads at 12.8 years, an all-time record.^[2] Passenger cars average 14.5 years, the oldest the category has ever been, which plants the typical sedan’s birthday somewhere around 2010 or 2011. That means the car parked in the average American driveway predates automatic emergency braking mandates, pedestrian detection systems, improved side-impact structures, and the lane-departure warnings that started appearing as standard equipment around model year 2015.^[3] Most of the 289-million-vehicle fleet missed the safety revolution entirely, and those vehicles are still commuting five days a week.

We ran the cross-tabulation ourselves, using a full decade of FARS fatal-crash data. Across 187,058 fatalities recorded from 2014 through 2023, pre-2015 model-year vehicles account for 154,631 deaths: 82.7% of the total.^[4] Vehicles built from 2015 onward, the cohort actually equipped with crash-avoidance technology that works in the real world, represent 17.3%. Even granting that older models accumulate more observation-years in the dataset, the concentration is extraordinary, and it points to a structural failure that no rate calculation can wallpaper: the safest cars in history sit in dealership lots and certified pre-owned listings while the vehicles doing 83% of the killing still roll out of suburban garages every morning.

Pre-2010 model-year vehicles alone account for 122,326 of those deaths, representing 65.4% of the total despite constituting a rapidly declining share of active registrations.^[4] A 2015 cutoff is generous, because many manufacturers did not make automatic emergency braking standard until 2018 or 2019, meaning the real denominator of technologically adequate vehicles on American roads is even smaller than 17.3%. When the IIHS tightened its 2026 Top Safety Pick criteria to require standard AEB for the first time, it was conceding what a decade of fatal-crash data had already proved: voluntary adoption and natural fleet turnover are not retiring dangerous vehicles fast enough to matter.^[3]

Fairness demands the counterargument at full volume, and it is a legitimate one. A per-mile fatality rate IS a valid metric; if thirty billion additional miles of driving produced 2,614 fewer deaths than the previous year, that reflects genuine per-trip safety improvement regardless of what the absolute count does.^[1] Population growth means more drivers, more drivers mean more miles, and the absolute death toll becomes mechanically harder to push down when total exposure climbs every single year. NHTSA enforcement partnerships, state-level DUI crackdowns, distracted-driving campaigns: these interventions save real lives, and dismissing the rate decline entirely would itself constitute bad statistics. But statistics that are technically correct and functionally misleading deserve scrutiny of their own, and America still kills more people per capita in traffic than virtually any peer nation.^[5]

Our FARS cross-tabulation carries real caveats that we should state plainly: it inherently overweights older model years because a 2005 car accumulates deaths across all ten observation years while a 2022 model gets barely two, and we cannot isolate fleet age from confounders like driver income, road type, or seatbelt compliance. That 2025 fatality estimate will eventually be revised, perhaps upward. None of those caveats erases what the data actually shows, which is that in terms of crash avoidance, the gap between a 2012 Camry and a 2024 Camry is not incremental but generational, and 82.7% of the people dying are stuck on the wrong side of it.

If your car was built before 2015, it lacks the collision-avoidance systems that define the divide between 82.7% and 17.3%, and the single most useful thing you can do right now is go to nhtsa.gov/recalls and enter your VIN. Open recalls on older vehicles remain the cheapest safety intervention available to any American driver. If you are shopping for a used car, prioritize model year 2018 or newer with standard AEB and a good IIHS rating, because the model year printed on the doorjamb sticker matters more than any option package on the window sticker.^[6] And if you want the fatality rate to mean something beyond actuarial theater, advocate for fleet renewal incentives, because the safest car on the road is currently the one most Americans cannot afford to buy.