Since July 2021, automakers have been legally required to tell federal regulators every time a vehicle running a driver assistance system crashes, and the resulting dataset is blunter than most car marketing ever gets. Of the 2,359 Level 2 driver-assistance crashes reported under NHTSA’s Standing General Order and tracked cumulatively by the nonprofit Center for Auto Safety, 2,030 involved Tesla vehicles running Autopilot or Full Self-Driving. That works out to 86 percent of every Level 2 crash logged, across every automaker combined. The gap between Tesla’s numbers and everyone else’s isn’t a rounding error, and it shows up whether you pull NHTSA’s own published summary reports or the raw filings researchers pull straight from the agency’s database.
What the Federal Order Actually Requires
NHTSA created the Standing General Order in 2021 to force real-world transparency on crashes tied to Level 2 driver assistance and fully automated driving systems. Manufacturers and fleet operators must report a redacted VIN, the vehicle’s make and model, a narrative of the crash, and whether the assistance system was active in the seconds before impact. A third amendment effective June 16, 2025 tightened the clock further: the most severe crashes are now due within five days, while lesser incidents get rolled into a monthly filing. NHTSA compiles the results into public CSV files updated every month, which is why outside researchers can build a running tally instead of waiting on the agency to editorialize the numbers.
Tesla’s Share Dwarfs Everyone Else
The 86 percent figure holds up across multiple points in the dataset’s history, and it’s not simply because Tesla sells more cars than rival brands with driver assistance features. Tesla vehicles report crashes automatically over cellular connection the moment telemetry detects an impact, while several competitors still lean on dealer service records or owner phone calls to generate a report. NHTSA’s own June 2022 summary report (DOT HS 813 325) says this outright, cautioning that crash totals “have not been normalized or adjusted by any measure of exposure” and shouldn’t be read as a direct safety ranking between brands. A car that reports every fender-bender automatically will always outnumber one that only shows up in the data when a service advisor remembers to file the paperwork.
That caveat matters more going forward. Under the same third amendment, Tesla became exempt from reporting an estimated 240 tow-away crashes with no injuries, fatalities, airbag deployment, or vulnerable road users involved, a change the Center for Auto Safety pegs at roughly 12 percent of Tesla’s historical total. The dataset that made Tesla’s dominance so visible is, going forward, going to undercount the very company that built it.
Where the Rest of the Field Shows Up
NHTSA’s June 2022 report, which tallied 392 Level 2 ADAS crashes reported as of May 15, 2022, named Tesla, Honda, and Subaru as the three manufacturers with the most crashes on record at that point. Honda Sensing and Subaru EyeSight are standard or near-standard equipment on some of the highest-volume vehicles on American roads, from the Civic to the Outback, so a bigger crash count there tracks with a bigger fleet running the hardware every day, not necessarily a worse system. It’s the same normalization problem NHTSA flags with Tesla, just running in the other direction.
Why a Tiny Crash Count Can Still Be the Bigger Story
Raw volume isn’t the only lens worth using. NHTSA opened investigation PE24012 into Ford BlueCruise on April 25, 2024, after just two crashes, both involving stationary vehicles sitting in active travel lanes on controlled-access highways at night. Those two incidents killed three people, across an estimated population of 130,050 equipped 2021–2024 Mustang Mach-E vehicles. Two crashes will never move a national percentage, but three deaths from a system engaged in both of them is exactly the kind of signal a pure ranking by crash count would bury.
Put the pieces together and the Standing General Order tells you two different things depending on which number you’re staring at. The company-by-company crash totals mostly measure how aggressively a system reports itself and how large its equipped fleet has grown, not which software is more dangerous per mile driven. The individual investigations, like the one into BlueCruise, are where severity actually lives. As more automakers push Level 2 and Level 3 systems into higher-volume models and NHTSA’s own reporting exemptions start trimming what shows up in the monthly file, the next few years of this dataset are going to require reading a lot more carefully than just scanning for the biggest number at the top.