Driver Deaths Fell 5%. Passenger Deaths Fell 12%. The Car Can't Save You From Yourself.

I ran the numbers, then I ran them again, and they didn't get better. NHTSA just released its sub-category breakdown of 2025 traffic fatalities, and buried in a table on page 4 is a statistic that should reframe every conversation about vehicle safety technology: passenger fatalities dropped 12 percent last year while driver fatalities dropped only 5 percent.^[1]

That is not a rounding artifact. Total traffic fatalities fell to an estimated 36,640 in 2025, a 6.7 percent decline from the 39,254 deaths reported in 2024. The overall fatality rate dropped to 1.10 per 100 million vehicle miles traveled, down from 1.19, and by any aggregate measure American roads got meaningfully safer last year. But aggregate measures lie by averaging, and when you split the data by who was sitting where, the improvement is radically unequal.

The sub-category data paints a coherent picture of which safety mechanisms are working and which problems they structurally cannot reach. Rollover fatalities dropped 11 percent, ejections fell 13 percent, unrestrained occupant deaths declined 13 percent, speeding-related fatalities fell 11 percent, and newer-vehicle (under 10 years old) occupant deaths dropped 12 percent. Urban arterial fatalities declined by 12 percent as well.^[1] Every category where technology can intervene between a threat and a body shows a double-digit decline. Electronic stability control prevents rollovers before they start, automatic emergency braking shaves speed on urban arterials before impact, and curtain airbags paired with improved side structures catch occupants that older designs would have ejected.

Now look at the categories where the threat originates inside the driver's seat. Rural interstate fatalities went up 1 percent, the only road type to increase. The 65-and-older age group saw zero improvement while every other age bracket declined, the 25–34 group dropping a full 13 percent.^[1] Pedalcyclist fatalities rose 4 percent, a problem driven largely by driver inattention at intersections where the cyclist never had a chance.

The pattern is not subtle, and it sharpens when you look at where improvements stalled. Modern vehicles have become extraordinarily good at protecting you from someone else's mistake. AEB reduces rear-end collisions, ESC keeps you on the road when a tire blows or a curve surprises you, and side-impact standards with curtain airbags mean a T-bone at 35 mph is survivable in ways it was not fifteen years ago. If you are a passenger, these systems work in your favor across the board because you are, by definition, not the person making the decisions that precede the crash.

Drivers sit in a different statistical universe. FARS toxicology data from 2014 through 2023 shows that one in five drivers involved in fatal crashes tested positive for alcohol or drugs: 15.1 percent for alcohol, 8.7 percent for drugs, 20 percent for any substance across nearly 491,000 tested drivers.^[2] Impairment is, almost exclusively, a driver problem. Passengers do not blow a .14 and decide to take the on-ramp at 90. They do not text through a red light. They do not fall asleep on a rural two-lane at 2 a.m. with their high beams off. These are the crash modes that single-vehicle crash data (down only 7 percent in 2025, versus 6 percent for multi-vehicle) and roadway departure data (down 9 percent) expose as the residual killing floor that technology keeps bumping against.^[1]

There is a legitimate counterargument, and it should be stated at full strength: the 12 percent passenger decline could partially reflect fewer passenger-miles driven, not a genuine per-exposure improvement. If average vehicle occupancy dropped, passenger exposure drops with it, and the fatality count falls even without any safety gain. NHTSA does not publish passenger-specific VMT, so the rate cannot be isolated. On the other hand, the same objection does not explain the direction of the gap. Driver VMT rose in 2025 (total VMT increased, per FHWA estimates), so if anything the driver denominator grew while the passenger denominator may have shrunk. That makes the 5 percent driver improvement, already the smaller number, even more remarkable.

An optimist reads this report and sees technology winning a war, because every crashworthiness metric is trending down. If every car on the road were a 2024-or-newer model with ESC, AEB, curtain airbags, and a reasonable hood height, the annual death toll would plummet. But it would not hit zero, because approximately one in five of the drivers behind the wheel in a fatal crash chose to be there impaired, and no sensor package fixes that choice after the key turns. The next wave of meaningful reduction requires going after the driver: ignition interlocks, phone-lockout systems, intelligent speed adaptation, or the blunt demographic force of autonomous vehicles displacing the human operator entirely. The 5-versus-12 gap is not a flaw in the data but the boundary between engineering and behavior, measured in bodies.

If you drive sober, belted, and attentive, the 2025 data says your odds improved by roughly double digits last year. If you do not, the car is doing everything it can, and the numbers say it is not enough.

Source: NHTSA FARS 2014–2023 and NHTSA CrashStats DOT HS 813 829. Sub-category estimates are ratio-adjusted projections based on partially coded 2025 FARS data; final counts may differ. Alcohol impairment data has significant reporting lag and is excluded from the 2025 projections. See methodology for caveats.