The Acceleration Problem: Are AV Crashes Getting Worse?

In 2025, NHTSA recorded 593 AV/ADAS crashes with known dates — a 51% increase from the 392 reported in 2024. At first glance, this acceleration is alarming. Are autonomous systems getting less safe? Are we deploying technology that crashes more often as it scales?

The answer, like most things in AV safety, is more nuanced than the headline suggests. To understand whether the crash rate is increasing, we need to account for two confounding variables: fleet size and miles driven.

Absolute Numbers vs. Rates

Between 2021 and 2025, the number of vehicles operating with AV/ADAS technology grew from an estimated 250,000 to 1.6 million — a 6.4x increase. Vehicle miles traveled by these systems grew even faster, from 3.2 billion to 52 billion — a 16x increase. Meanwhile, crashes went from 90 to 593 — a 6.6x increase.

When we normalize crashes by VMT, the rate actually fell from 2.81 per 100M VMT in 2021 to 1.14 in 2025 — a 59% improvement. The crash rate per 1,000 vehicles tells a similar story, hovering between 0.36 and 0.42 across all five years. More vehicles are crashing in absolute terms, but each vehicle (and each mile) is getting safer.

The 2025 Anomaly

There is one concerning data point: 2025 is the first year where crash growth (51%) outpaced fleet growth (45%). In every prior year, fleet expansion explained most or all of the increase in absolute crashes. In 2025, something else may be contributing.

Several hypotheses could explain this: (1) Rapidly deployed new systems from emerging manufacturers may have higher initial crash rates. (2) Expansion into new geographic markets means encountering unfamiliar driving conditions. (3) Increased reporting compliance means we're simply catching more incidents that previously went unreported. (4) Systems operating in more complex environments (urban FSD, for instance) face more challenging conditions than highway-only operation.

The VMT Story Is More Positive

The crash rate per 100M VMT paints a more optimistic picture because VMT growth has consistently outpaced fleet growth. This means each vehicle is driving more autonomous miles — a sign of increasing trust and usage. When crashes per VMT falls even as total crashes rise, it means the technology is genuinely improving on a per-mile basis.

The 2025 rate of 1.14 crashes per 100M VMT compares favorably to the general US crash rate of approximately 2.07 per million VMT (which translates to roughly 207 per 100M VMT for police-reported crashes). However, this comparison is imperfect — the SGO reporting threshold differs from police crash reports, and AV operating conditions differ from general driving.

Cumulative Weight

By the end of 2025, the cumulative total of AV/ADAS crashes with known dates reached 1,544. (The NHTSA database contains an additional 4,624 incidents with unknown dates, many of which are Tesla ADAS complaints that couldn't be precisely dated.) The cumulative curve is accelerating — not because the rate is increasing, but because the denominator (fleet × miles) keeps growing exponentially.

Looking Ahead: 2026 Projections

Based on current fleet growth trajectories and the observed crash rate trend, we project approximately 850 AV/ADAS crashes in 2026, with a range of 650-1,100 depending on fleet expansion and rate improvement. If the crash rate continues to fall at its current pace, the number could be on the lower end. If fleet growth accelerates (as expected with Waymo's expansion and Tesla's FSD rollout), absolute numbers will climb even as rates improve.

The projection underscores a fundamental tension in AV safety communication: absolute crash counts will keep rising for years, potentially decades, even as per-mile safety improves. The public, media, and regulators will need to become literate in rate-based thinking — or risk making policy decisions based on misleading trend lines.

The Bottom Line

Are AV crashes getting worse? In absolute numbers, yes — dramatically. In crash rates, no — they're improving. Both statements are simultaneously true, and both matter. The challenge for the industry is that headlines run on absolute numbers, while safety engineers think in rates. Bridging that gap — with transparent, rate-based reporting — is essential for maintaining public trust as the technology scales.