What Tesla can show
A large, connected fleet lets Tesla compare collisions and mileage across control types using the same telemetry pipeline. Its report updates on a rolling 12-month basis.
Last reviewed August 14, 2026
Tesla FSD Safety Statistics 2026: What Tesla Reports—and What Independent Data Can Tell Us
Tesla now publishes a detailed collision-frequency report for FSD (Supervised). It is useful evidence—yet it is not an independent, fault-based, injury, or fatality rate.
The honest answer to “Is FSD safer?” is more specific: Tesla’s own telemetry shows one kind of comparison; public regulators and independent research answer different questions.
Written
Updated
The short answer
What the public data cannot show
There is not yet a public, independently normalized FSD rate for crashes, injuries, fatalities, fault, road type, weather, software version, and matched driver behavior.
What drivers should do
Treat FSD as supervised assistance. A driver must remain engaged and ready to take control; its name is not a safety permission slip.
What Tesla’s report actually measures
Read a chart label before comparing a headline number. Tesla’s current report is a collision-frequency report—not a tally of who caused a crash or who was injured.
| Metric | Tesla’s published definition | Why it matters |
|---|---|---|
| Major collision | Collision with an airbag or other non-reversible restraint deployment. | It is a severity proxy, not an injury or fatality count. |
| Minor collision | At least an 8 km/h delta-V event within 150 milliseconds without restraint deployment. | It uses a telemetry threshold; it is not the same thing as every police-reported crash. |
| FSD engaged | FSD was active at any point in the five seconds before the collision event. | That includes some cases where the driver or system disengaged shortly before impact. |
| Fault | Tesla does not assign it in the report. | A lower collision frequency is not proof that the system caused or prevented every event. |
Source: Tesla’s FSD (Supervised) Vehicle Safety Report methodology.
Why Tesla’s comparison is informative but not the final word
A strength: common instrumentation
Tesla records mileage and collision telemetry from the same fleet, then separates FSD (Supervised), manual driving with active safety features, and manual driving without them. That is more internally consistent than comparing disconnected manufacturers’ raw crash totals.
A limit: users and roads are not randomized
FSD use can vary by driver, route, weather, time of day, traffic, vehicle, software version, and the moments when people choose to activate it. A within-fleet comparison reduces some measurement differences; it does not turn an observational comparison into a randomized trial.
A limit: collision frequency is not every safety outcome
Tesla says its telemetry report does not measure injury data and does not determine fault. It therefore cannot, by itself, answer a question about injury risk, fatalities, vulnerable road users, or legal responsibility.
How to read NHTSA’s FSD and ADAS crash reports
NHTSA’s Standing General Order is valuable for surfacing incidents and potential defects. It is not a manufacturer leaderboard.
Raw totals are not rates
NHTSA explicitly says its public incident reports are not normalized for vehicles in service, miles traveled, operational design domain, or exposure.
Reporting visibility differs
A highly connected fleet may learn about more crashes than a manufacturer relying mainly on consumer reports. Multiple reports can also describe the same incident.
It still matters
The data can identify scenarios worth investigation. It should be read as a safety signal and case record, not divided into a simple “crashes per company” ranking.
Independent evidence: promising tools, unanswered system-level question
Independent research does not yet supply a Tesla FSD-specific, matched crash-rate answer. It does show why driver monitoring, safeguards, and careful study design matter.
IIHS: partial automation safety benefits remain unproven
A 2024 IIHS analysis of Nissan and BMW systems found no convincing evidence that partial driving automation is preventing crashes in the real world. It is not a Tesla FSD test, but it is a useful warning against treating a convenience feature as a proved safety system without exposure and system-use data.
IIHS: the tested Tesla safeguards were rated poor
IIHS separately evaluates how partial-automation systems keep drivers engaged through monitoring, attention reminders, emergency procedures, and other safeguards. Tesla Autopilot and FSD (Beta), both tested as software version 2023.7.10 in a 2021–23 Model 3, received poor overall safeguard ratings. These are dated, version-specific ratings of misuse protections—not a crash rate or a measure of how well current FSD performs the driving task.
NHTSA’s reduced-visibility evaluation is closed, not a blanket verdict
NHTSA opened PE24-031 after four reports of FSD-engaged crashes in reduced visibility, including one fatal pedestrian crash. The public record lists the preliminary evaluation as closed on March 18, 2026. That one-page closing record does not turn the four cases into a fleet-wide crash rate or establish a general safety conclusion.
Sources: IIHS partial-automation crash-rate study; IIHS partial-automation safeguard ratings; NHTSA PE24-031 closing record; NHTSA opening summary.
A practical checklist for any FSD statistic
- Ask what is counted. Is it a telemetry-defined collision, a police-reported crash, an injury, or a fatality?
- Ask what is in the denominator. Miles, vehicles, trips, road type, and time period all change the meaning of a rate.
- Ask whether the comparison is matched. Highway driving, city driving, driver choice, weather, and software version cannot be assumed equivalent.
- Separate system performance from driver responsibility. A supervised feature can help and still require immediate human intervention.
- Prefer a dated primary source. Tesla says its report updates quarterly; save the report date with any number you quote.
Related Tesla safety research
FSD statistics and vehicle crashworthiness are different questions. Review both before drawing a conclusion about a particular Tesla model.
Tesla crash tests and death rates Our methodology