Tesla's latest Vehicle Safety Report claims cars driving with FSD (Supervised) go 5.69 million miles between major collisions, versus 698,781 for the U.S. average. The headline numbers are real. The comparison behind them is where independent safety researchers say the picture gets a lot less clear-cut.
Tesla has updated its Vehicle Safety Report with new FSD (Supervised) collision data, covering a rolling twelve-month period across North America. The headline claim: Teslas driving with FSD (Supervised) engaged went an average of 5,690,675 miles between major collisions — airbag-deployment-level crashes — compared to 698,781 miles for Tesla's estimated U.S. average. Tesla frames that as roughly 7x fewer major collisions with FSD engaged. The raw numbers are real and internally consistent. Whether that comparison tells you what Tesla wants it to tell you is a separate question, and independent safety researchers have been picking that question apart for months.
Tesla breaks its data into four groups: vehicles driving with FSD (Supervised) engaged, Teslas driven manually with active safety features (Automatic Emergency Braking, Forward Collision Warning, etc. — standard on every Tesla made since 2014), Teslas driven manually with no active safety features (pre-2014 Tesla vehicles, which Tesla uses as a proxy for an older, less-equipped car), and an estimated U.S. average built from federal crash and mileage data.
| Group | Miles per Major Collision | Total Major Collisions | Total Miles Logged |
|---|---|---|---|
| FSD (Supervised) engaged | 5,690,675 | 1,255 | 7,141,797,525 |
| Manual, with active safety | 2,076,014 | 17,380 | 36,081,121,994 |
| Manual, no active safety | 859,465 | 239 | 205,412,073 |
| U.S. average (estimated) | 698,781 | 4,646,404 | 3,246,817,000,000 |
A "major collision" in this report specifically means a crash severe enough to deploy an airbag or other non-reversible restraint — a fairly high bar. Tesla also counts a collision as an "FSD collision" if the system was active at any point in the five seconds before impact, even if the driver had already taken over. The report only covers North America; Tesla publishes separate figures for other regions and for non-highway driving specifically.
The 7x figure comes from comparing FSD-engaged miles to the "U.S. average" column — which Tesla itself defines as pre-2014 vehicles, standing in for the average American car (roughly 12-13 years old) because that's genuinely closer to the actual U.S. fleet age than a new Tesla. That's a defensible choice for estimating national road safety broadly, but it also means the comparison is measuring a brand-new car loaded with modern collision-avoidance hardware against a car built before most of that hardware existed — a gap that has nothing to do with whether FSD itself is doing anything.
If the question is whether turning FSD on makes a Tesla safer than driving the same Tesla manually, the more relevant baseline in Tesla's own table isn't the U.S. average — it's the "manual, with active safety" row, since that's the same fleet, same hardware, same drivers, with only the FSD variable changed.
Run that comparison instead — 5,690,675 miles per collision with FSD versus 2,076,014 miles per collision for the same fleet driving manually — and the gap narrows to roughly 2.7x, not 7x. That's still a real difference, but it's a much smaller and more defensible one, and it's the comparison that Carnegie Mellon University engineering professor Phil Koopman, who has reviewed multiple versions of this report, argues is the only apples-to-apples one available in Tesla's own data.
A 2026 Reuters investigation into Tesla's FSD safety claims — drawing on interviews with former Tesla data labelers, a former self-driving engineer, and eleven independent traffic-safety researchers — found similar issues with earlier versions of this same report, concluding that stacking the fleet-age gap, the crash-type mismatch, and selection-and-exposure bias together could inflate Tesla's headline comparison by somewhere close to an order of magnitude versus a properly matched baseline. Ten of the eleven researchers Reuters consulted described Tesla's safety statistics as marketing material rather than a rigorous safety analysis.
Separately from the collision-rate debate, Tesla also updated its FSD mileage tracker on July 22 to reflect a larger vehicle fleet and rising usage. The company now says its fleet is driving an average of 37.5 million miles per day with FSD engaged — up from 28.8 million miles per day just a few months ago, and more than double the 14.4 million miles per day reported earlier in 2026. At that rate, Tesla's fleet logs 1,000 FSD miles roughly every 2.3 seconds. Annualized, that works out to about 13.7 billion FSD miles a year, up 241% year over year.
This growth curve matters for how to read the safety numbers above: a larger, more actively used fleet means Tesla's collision-rate denominators are built on a bigger and more current sample than in past reports, which is a genuine methodological improvement even as the framing critiques above still apply. It also means whatever safety profile FSD has — good, bad, or somewhere in between — is now touching a meaningfully larger share of daily American driving than it was even a few months ago.
On the same day, Tesla separately touted its driverless Robotaxi network specifically, posting that it has recorded "0 notable incidents across over 380,000 miles traveled by Robotaxi."
Worth noting: "notable incidents" is Tesla's own term and isn't defined the same way "major collision" is elsewhere in the Vehicle Safety Report, so it's not directly comparable to the airbag-deployment threshold used in the collision-rate data above. It's also a much smaller mileage sample — 380,000 miles is a rounding error next to the 7.1 billion FSD miles in the main report — which makes a claim like "zero incidents" easier to sustain over a short window regardless of the technology's long-run safety profile. Worth watching whether that number holds up as the Robotaxi network's mileage grows, particularly given the fleet-size and crash-rate concerns already raised about the Robotaxi network in Austin, Dallas, and Houston elsewhere on this site.
None of this means FSD (Supervised) is unsafe, and it's worth being precise about what the criticism actually targets: it's aimed at the size and framing of Tesla's comparison, not at whether the underlying collision counts are fabricated. Even the more conservative same-fleet comparison (roughly 2.7x) is a meaningful gap, and it's consistent with the broader industry view — including from the Insurance Institute for Highway Safety — that modern driver-assistance systems generally correlate with fewer severe crashes when a driver is still supervising the car and cross-traffic conditions remain constant across the comparison.
Where the debate actually lives is in whether "7x safer" is a fair characterization for a general audience, given how much of that multiplier comes from comparing a new car to a 12-year-old one rather than from FSD's own contribution. Regulators appear to be asking the same question — Tesla's European FSD approval push has drawn direct scrutiny over similar self-reported statistics, and the system remains the subject of multiple active NHTSA investigations in the U.S., unrelated to this specific report.
The bottom line: Tesla's raw numbers — 5.69 million miles per major collision with FSD engaged, versus roughly 2.08 million for the same fleet driving manually — represent a real and fairly substantial gap. The "7x" headline figure is technically accurate as a description of the U.S.-average comparison Tesla chose to publish, but it isn't the most informative number in Tesla's own table, and independent researchers have consistently argued the same-fleet comparison is the one that actually answers the question most readers assume "7x safer" is answering.
Tesla's Q2 2026 earnings show a company spending heavily and earning less for it: non-GAAP EPS missed estimates at $0.33, free cash flow went negative, and gross margin fell to 16.8%. Meanwhile Cybercab began production, Optimus took over the Fremont lines once used for Model S/X, FSD subscriptions grew faster than ever, and Robotaxi passed 2.4 million paid miles.
Tesla raised lease pricing on every Model 3 trim in the US on July 21, 2026, with the base RWD jumping 15.2% to $379/month. It's the second directional shift on Model 3 leases in a single quarter, coming just three months after Tesla cut pricing to some of the lowest rates the car has ever seen.
Tesla published the release notes for its 2026 Summer Update on July 21, and it's a software-focused drop rather than a hardware one: Grok can now make phone calls and open your glovebox, the mobile app gets shareable self-driving stats, Caraoke scores your singing, navigation gets smarter about your routine, and you can now upload custom wraps from your phone instead of a USB drive.