GM Says Its Safety Tech Prevents 86% of Crashes. The Math Saves 3%.

On May 8, General Motors and the University of Michigan Transportation Research Institute published a study spanning 12 million vehicles and 700,000 police-reported crashes across 18 states, the largest real-world evaluation of advanced driver assistance systems ever conducted on a single manufacturer's fleet.^[1] The headline number was irresistible: rear cameras and cross-traffic alerts reduce crashes by 86 percent. Every outlet ran it. Nobody did the subtraction.

Eighty-six percent of what, exactly? Backing crashes. Reversing into a shopping cart corral or, tragically, a child standing behind a driveway. NHTSA data shows backing incidents account for roughly 200 to 300 fatalities per year out of approximately 38,000 total.^[2] Less than one percent of the body count. GM built a technology that is spectacularly effective at preventing one of the rarest ways Americans die on roads, then let the percentage do the talking while the other 37,700 deaths stood politely off-camera, waiting for someone to notice they were not invited to the press release.

GM's study reported other findings that deserve examination. Forward automatic emergency braking cut rear-end crashes by 57 percent. Pedestrian AEB reduced pedestrian injury crashes by 35 percent. Lane keep assist lowered lane-change crashes by 14 percent.^[1] All real numbers from a legitimate study with a genuinely massive sample size, and all of them carefully measured against police-reported crashes rather than fatal ones. That distinction is not a footnote; it is the entire story.

The Ceiling Calculation Nobody Published

Map GM's own reduction rates onto NHTSA's fatal crash type distribution and you get a theoretical lives-saved ceiling. Rear cameras at 86 percent effectiveness against backing fatalities: 258 lives. Forward AEB at 57 percent against rear-end fatalities, which constitute about 7 percent of the annual total: 1,516 lives. Pedestrian AEB at 35 percent against pedestrian deaths running around 20 percent of the total: 2,660 lives. Lane keep assist at 14 percent against the 4 percent of fatalities from sideswipe and lane-change crashes: 213 lives.^[2][3]

Grand total at full fleet penetration, assuming every car on every road has every feature working perfectly: approximately 4,647 lives per year. That is 12.2 percent of the annual death toll, which is substantial but a long way from 86.

But full fleet penetration does not exist. Model year 2020 and newer vehicles represent roughly 25 to 30 percent of registered vehicles, and in FARS fatality data they account for just 3.2 percent of deaths because newer cars are inherently safer and have accumulated less exposure time.^[2] At current penetration the ceiling drops to around 1,162 lives. Discount further because GM measured police-reported crashes while fatal crashes involve higher speeds where AEB merely reduces closing velocity rather than preventing impact, and the realistic estimate settles between 581 and 1,162 lives saved annually. Call it 1.5 to 3 percent of the 38,000.

Fifty-three percent of fatal crashes are single-vehicle events: run-off-road, rollovers, fixed-object impacts, the kind of catastrophic loss-of-control events that happen at speeds where lane keep assist at 14 percent effectiveness is a rounding error in the morgue's paperwork.^[3] Not a single technology in GM's study addresses them meaningfully. Head-on collisions kill another 10 percent of the total, and they are also absent from this analysis. Every feature in GM's ADAS suite was designed for parking lots and highway following distances. Where people actually die at scale was never the target.

What You Should Do

If you drive a pre-2020 vehicle, none of these technologies protect you, and your car represents the overwhelming majority of the fleet that is still killing people. Your single most effective safety upgrade is buying a newer vehicle with standard ADAS, because GM's study does prove the technology works within its domain. If you already drive a 2020-or-later GM vehicle, understand what the systems can and cannot do: forward AEB is tested at closing speeds under 40 mph, pedestrian AEB under 25 mph, and neither system is designed to prevent a 65-mph head-on collision. Calibrate your trust accordingly and check your VIN for open recalls at nhtsa.gov/recalls. And when you see a headline claiming any technology reduces crashes by 86 percent, ask the question the headline is betting you will not ask: 86 percent of which crashes?

Limitations

This analysis cross-references two fundamentally different datasets. GM-UMTRI measured police-reported crash reductions across all severity levels; FARS captures only the approximately 38,000 annual fatal crashes, which represent a fraction of the roughly 6.7 million total U.S. crashes per year. A technology that dramatically reduces fender-benders may show modest fatal-crash impact without that being a failure. Fatal crash type percentages used here are approximate, drawn from NHTSA Traffic Safety Facts rather than our own FARS dataset, which lacks crash-type coding at that granularity. The 50 percent discount applied to estimate fatal versus police-reported effectiveness is an assumption, not a measurement; actual AEB performance at lethal speeds could be higher or lower. Fleet penetration figures are approximate because no definitive count of ADAS-equipped vehicles on U.S. roads exists.

The Counterargument

GM would fairly point out that fleet turnover is a trajectory, not a snapshot, and that the 12.2 percent ceiling at full penetration is still nearly 5,000 lives per year, which would rank among the most significant automotive safety interventions since seatbelts. They would add that injury crash prevention matters enormously even when it does not appear in fatality statistics: the 700,000 crashes in their study include hospitalizations, permanent disabilities, and economic costs that FARS does not capture. The 86 percent reduction in backing crashes saves a disproportionate number of children and elderly pedestrians, which is a population-adjusted impact far larger than the raw percentage suggests. And this is one manufacturer's study of one generation of technology; industry-wide deployment across all OEMs, combined with improving sensor capabilities and eventually vehicle-to-vehicle communication, could expand the addressable share of fatal crashes well beyond what current ADAS targets. All of this is valid, and the counterargument is about trajectory. The headline was about today.