36,640 People Died on American Roads in 2025. The Cars Didn't Get Safer.

In March, we ran the numbers on a single month that dipped below the pre-COVID fatality rate and warned you not to get used to it. Our full-year estimate: 37,000 to 39,000 deaths. We were wrong. The actual number is 36,640.

We missed low by at least 360 bodies. In traffic safety, being wrong in this direction is the kind of error you frame on the wall.

NHTSA released the 2025 full-year estimate alongside final 2024 data last month. The trajectory reads like a controlled descent: 42,721 deaths in 2022, 40,901 in 2023, 39,254 in 2024, and now 36,640 in 2025.^[1][2] A 14.2 percent drop in three years. Four consecutive years of decline. The rate of 1.10 fatalities per 100 million vehicle miles traveled is below the 2013-2019 average of 1.13 for the first time on a full-year basis.^[3]

Americans also drove more. VMT climbed 29.8 billion miles, about 0.9 percent, meaning more driving produced fewer deaths. Fatalities dropped in 39 states, the District of Columbia, and Puerto Rico.

So what changed?

Not the cars.

The fleet didn't move

The average vehicle on an American road is 12.6 years old. The Chevrolet Silverado that killed at a rate of 1.25 per 100 million VMT between 2014 and 2023 is still out there, still being sold, still the same basic platform. The Ford F-150 at 1.04, the Honda Accord at 3.07, the Ford Mustang at 6.02.^[4] None of these vehicles were redesigned between 2024 and 2025. No new federal safety mandate took effect. The AEB requirement for passenger cars does not arrive until 2029.^[5] Electronic stability control, the last major mandate, was locked in by 2012 and has been fully absorbed into the fleet for years.

The hardware did not change. The software running it did. And by software, I mean us.

2,614 people saved by behavior alone

If the same fleet killed 39,254 people in 2024 and 36,640 in 2025, the 2,614-death gap is almost entirely behavioral. Some of that is enforcement: increased DUI checkpoints, speed camera expansion in a handful of jurisdictions, distracted driving campaigns that may have finally landed. Some is economic: fuel prices, employment patterns, and commuting habits shift driving exposure in ways that VMT alone does not capture. A shorter commute at 35 mph replaces a longer one at 65. Same miles, different risk.

But the mechanical implication is stark. Per-vehicle fatality rates in our FARS data measure a floor, not a prediction. The Impala's rate of 5.0 per 100M VMT is what that vehicle does under the full spectrum of American driving behavior. When behavior improves, the actual death count falls below what the hardware would predict. When behavior degrades, as it did during COVID, the body count surges above the hardware baseline by thousands. Between 2020 and 2024, that behavioral overshoot killed an estimated 21,500 people above the pre-pandemic trend.

What this means for you

Two things.

First: the vehicle you drive still matters enormously. A 2006 Chevrolet Cobalt and a 2024 Toyota RAV4 exist in different safety universes regardless of how soberly you drive either one. Fleet turnover is gradual, and replacing old iron with modern crash structures, automatic emergency braking, and lane-keeping remains the single highest-impact upgrade available. Check your VIN at nhtsa.gov/recalls.

Second: behavior is a multiplier, not an additive. The same vehicle driven drunk at 85 mph and sober at 60 mph occupies two entirely different risk profiles. The 2025 data proves that shifting behavior across the population saves more lives per year than any single vehicle mandate in the pipeline. The 2,614-person gap between 2024 and 2025 is larger than the entire body count of any individual vehicle model in our dataset except the Silverado and the F-150.

Strongest counterargument

Fleet turnover is continuous, not zero. Even without a mandate, every new car sold in 2025 replaced something older, and the average new vehicle carries better ADAS than whatever it retired. Forward collision warning, automatic emergency braking, blind-spot monitoring, and adaptive headlights are now standard on most new models. Over a fleet of 290 million vehicles, even small annual improvements compound. Attributing the entire 2,614-death decline to behavior is an overstatement; the engineering contribution is small per year but real. Additionally, trauma medicine improvements may be converting some would-be fatalities into severe injuries, which flatters the death count without making the road any safer.

Limitations

The 2025 figure of 36,640 is NHTSA's early estimate, subject to revision of 2-4 percent in either direction. VMT figures from FHWA are modeled, not measured. Our FARS per-vehicle data covers 2014-2023 and cannot yet confirm whether individual model rates shifted in 2024-2025. The behavioral vs. engineering decomposition is our original calculation based on the absence of major fleet changes, not a controlled analysis. State-level data is preliminary; the 39-state decline and 10-state increase may shift with final reporting.