Autonomous Vehicles Crash Twice as Often But Prove Safer Overall — 9.1 vs 4.1 Per Million Miles

Self-driving cars crash more frequently than human drivers, but their accidents are less likely to be fatal. A new reality is emerging from the data: more crashes doesn't necessarily mean more danger. The National Law Review reports that fully autonomous and semi-autonomous vehicles have 9.1 crashes per million miles driven, compared to 4.1 for regular cars. Yet only 2 people have died in fully autonomous vehicle accidents as of March 2026, while semi-autonomous systems like Tesla Autopilot have been linked to 55 deaths.

The difference reveals something important about autonomous vehicle safety: frequency of crashes and severity of crashes are not the same thing.

Why Autonomous Vehicles Crash More

Autonomous vehicles are more sensitive to certain types of incidents that human drivers might handle instinctively. They encounter edge cases and unusual situations that don't fit their training data perfectly. A child running into the street from behind a parked car, an unusual traffic pattern, or adverse weather conditions can trigger accidents that a human driver might avoid through experience and intuition.

Read Also: EV Charging Stations Fail 1 in 5 Attempts — Only 71% Success Rate Creates "Charging Anxiety"

Waymo, the leading autonomous vehicle operator, conducted research comparing its safety record to human benchmarks. The company reported a 57% reduction in police-reported crashes compared to human drivers. Waymo's accident rate stands at 2.1 incidents per million miles driven, while human drivers average 4.85 incidents per million miles — making Waymo's self-driving technology 2.3 times safer by that metric.

But that data point — Waymo's advantage — competes with the 9.1 vs 4.1 national comparison. The difference comes down to what's being measured and how crashes are categorized.

The Fatality Gap

Here's where the real safety story emerges. Fully autonomous vehicles have caused only 2 deaths as of March 2026, despite years of operation across multiple companies. Semi-autonomous systems — cars with driver assistance features like Tesla's Autopilot and Full Self-Driving — are linked to 55 deaths. Tesla's Q1 2023 safety report claimed 1 crash per 5.18 million miles using Autopilot.

Not sure which car to choose? Take our quiz and find out!

The distinction matters because semi-autonomous systems require human supervision that isn't always present. A driver using Autopilot might be distracted, sleeping, or simply overconfident in the technology's capabilities. When accidents happen, the human should theoretically intervene, but often doesn't.

Fully autonomous vehicles are different. There's no expectation that a human will take control. The system is responsible for safety in all driving situations. That accountability appears to be producing measurable results in terms of fatalities, even if crash frequency is higher.

California's Autonomous Vehicle Reality

California reports the highest number of crashes involving autonomous vehicles, with over 1,600 incidents across both fully autonomous and semi-autonomous systems. The state's extensive autonomous vehicle testing program means more data, which reveals both successes and failures.

In January 2026, a Waymo vehicle struck a child in Santa Monica during morning drop-off hours. The child ran into the roadway from behind a double-parked SUV, directly into the vehicle's path. The Waymo braked hard, slowing from 17 mph to under 6 mph before impact. The controlled response — a rapid but measured deceleration — exemplifies how autonomous systems handle unavoidable accidents: they minimize impact force even when collision is inevitable.

That's fundamentally different from human driver behavior in panic situations, where drivers often brake too late or with insufficient force.

What Americans Think

Public perception of autonomous vehicle safety remains divided. The AAA Foundation for Traffic Safety found that Americans' trust in self-driving cars improved between 2024 and 2025, with those expressing trust increasing from 9% to 13%. Progress is slow, but real.

When asked whether autonomous vehicles would impact traffic deaths and injuries, Americans split three ways: 31% felt there would be no impact, 39% thought autonomous vehicles would decrease death rates, and 27% believed the opposite. That 39% represents growing confidence that the technology could actually save lives.

The Waymo Benchmark

Waymo's safety record provides the strongest evidence that fully autonomous vehicles can outperform human drivers in critical ways. Over 7 million miles driven, the company reported only 3 high-severity collisions. The company also achieved what sounds impossible: a 100% reduction in crashes where the autonomous vehicle was the primary at-fault party over 1 million miles.

That doesn't mean Waymo vehicles never cause accidents. It means when accidents occur, they're caused by other factors — another driver's negligence, unpredictable pedestrian behavior, weather, road conditions. The autonomous system isn't failing and causing crashes; the environment is.

Looking Forward

The potential safety gains from widespread autonomous vehicle adoption are enormous. Advanced driver assistance systems (ADAS) could prevent up to 20,800 deaths per year in the US if deployed universally. Level 4 autonomous trucks could save 1,000 lives annually on US highways alone. Full autonomy could reduce traffic accident costs by $190 billion annually.

But those gains are still aspirational. The real-world data remains mixed and incomplete. The Association for Computing Machinery warned policymakers in April 2024 that "they should not assume that fully automated vehicles will necessarily reduce road injuries and fatalities."

The Bottom Line

Self-driving cars crash more often than human drivers but appear to cause fewer fatal accidents. This paradox suggests that as autonomous technology matures, crash frequency may decrease while safety improves. What matters isn't whether a vehicle gets into an accident — it's whether that accident kills someone.

For now, the data shows fully autonomous vehicles are proving safer. Semi-autonomous systems, which depend on inattentive humans to intervene, remain dangerous. The distinction will shape which autonomous technologies survive and which get replaced.