Tesla records collision events in accordance with 49 C.F.R. § 563.5, which defines a collision event as a crash or physical occurrence that causes any non-reversible deployable restraint to be deployed or that meets or exceeds the trigger threshold of a change in velocity (Delta-V) of 8 km/h occurring within a 150-millisecond period. Delta-V measures the change in a vehicle's velocity during a collision and is widely used in crash analysis to quantify impact severity. A higher Delta-V indicates greater forces acting on the vehicle and its occupants, making it a reliable indicator for discriminating between impact severities across a broad spectrum of collisions.

Consistent with 49 C.F.R. § 563.5, Tesla separates collisions into two categories:

1. Deployment collisions tend to have higher severity impacts where the vehicle's airbags or other non-reversible pyrotechnic restraints are deployed. Tesla refers to these collisions as “major collisions” in the Vehicle Safety Report.
2. Non-deployment collisions tend to have lower severity impacts (Delta-V ≥ 8 km/h within 150 milliseconds) where neither airbags nor any other non-reversible pyrotechnic restraints are deployed. Tesla refers to these collisions as “minor collisions” in the Vehicle Safety Report.

If FSD (Supervised) was active at any point within five seconds leading up to a collision event, Tesla considers the collision to have occurred with FSD (Supervised) engaged for purposes of calculating collision rates for the Vehicle Safety Report. This approach accounts for the time required for drivers to recognize potential hazards and take manual control of the vehicle. This calculation ensures that our reported collision rates for FSD (Supervised) capture not only collisions that occur while the system is actively controlling the vehicle, but also scenarios where a driver may disengage the system or where the system aborts on its own shortly before impact.

Tesla does not attribute fault in our collision reporting due to the subjective and complex nature of fault determination, which typically requires detailed investigation of circumstances, driver intent and applicable traffic laws. By reporting all collisions, regardless of who or what factors contributed to the cause of the collision—driver(s), vehicle(s), environment, system misuse, etc.—Tesla provides a consistent and objective measure that allows for meaningful safety comparisons across different driving modes and conditions.

Tesla compares FSD (Supervised) collision rates, observed by our telemetry, to both manually driven Tesla vehicles with and without active safety features. All Tesla vehicles manufactured after 2014 are equipped with active safety features, including Automatic Emergency Braking, Forward Collision Warning, Lane Departure Warning and other collision-avoidance systems. Tesla’s data reflects substantially higher miles-per-collision rates for Tesla vehicles equipped with active safety features compared to Tesla vehicles without these features.

Tesla vehicles without active safety features (i.e., Tesla vehicles manufactured prior to 2014) serve as the best proxy for the safety of an average U.S. vehicle, given that the average vehicle age in the U.S. is approximately 12 years old, predating widespread adoption of active safety and advanced driver-assistance systems. While these vehicles are not sold anymore, they collectively drive hundreds of millions of miles every year.

The Tesla fleet is comprised of millions of vehicles worldwide, generating real-world driving data at scale. The Vehicle Safety Report illustrates a combination of anonymous vehicle telemetry, including miles driven, road classifications and control type, in addition to safety-critical event telemetry (VIN-associated data received when the vehicle experiences a major or minor collision, as defined above). To illustrate telemetry volume, in Q3 of 2025 alone, Tesla received 2.5 billion telemetry packages from our vehicle fleet around the world (excluding China).

The vehicle telemetry packages that are analyzed and aggregated in the Vehicle Safety Report occur in two distinct ways:

1. On shift-to-park: The vehicle transmits a data packet containing anonymized mileage information from the culminating trip (or drive cycle) to Tesla servers. This data informs Tesla of the number of miles traveled in different control types and road classifications for each vehicle type.
2. On detection of a major or minor collision: A VIN-associated data packet is transmitted to Tesla servers. If the data upload fails, trip and collision information remains stored on the vehicle itself and is available to reattempt upload if connectivity is restored.

Certain variables, including but not limited to cellular connectivity or damage to vehicle communication systems, may affect Tesla’s ability to capture certain events. Importantly, these variables apply uniformly across all Tesla control types analyzed in the Vehicle Safety Report, preserving the validity of comparative safety metrics between FSD (Supervised) and manual driving.

Tesla categorizes roadways into highway and non-highway based on road characteristics including map data routing classes, lane count and controlled access infrastructure. While our classification accuracy is high, there may be minor variations in mileage and collision data based on localization, map dependencies and differences in definitions when compared to U.S. averages.

The safety benefits of FSD (Supervised) are clear when compared to manually driven Tesla vehicles with and without active safety features. This is the most direct and statistically valid comparison setup since it is being made within the same fleet of vehicles using the same telemetry pipelines.

Additionally, Tesla vehicles share the road with many road users in non-Tesla vehicles. To that end, it is prudent to assess the safety of FSD (Supervised) against the general safety of roadways in the U.S. A quantifiable means by which to do this is to estimate the U.S. collision rate using data published by the U.S. government, which is the best available source. To establish a baseline U.S. average, Tesla used U.S. government data, as explained below, to estimate for total miles traveled (numerator) and total vehicles involved in a collision (denominator).

The U.S. government publishes several data sources for consideration. For total miles traveled (numerator), a commonly used source for research is the Federal Highway Administration’s (FHWA) vehicle miles traveled (VMT) annual reports (with the most recent publication being for calendar year 2023). For total vehicles involved in a collision (denominator), Tesla looked to three sampling and reporting systems: the National Highway Traffic Safety Administration’s (NHTSA) Crash Report Sampling System (CRSS), which captures police-reported incident rates nationwide; second, NHTSA’s Crash Investigation Sampling System (CISS), which samples police-reported collisions where at least one passenger vehicle was towed from the crash scene; and third, NHTSA’s Fatality Analysis Reporting System (FARS), which measures fatality events nationwide. Among these sampling and reporting systems, the CISS primarily involves collision severities that most closely align with Tesla’s approach to tracking major collisions, because the CISS captures collisions where at least one passenger vehicle was towed and is not limited by injury outcomes. In contrast, CRSS primarily captures incidents where neither airbags nor other restraints may have deployed (e.g., approximately 71.7% of crashes in the CRSS for 2023 are property damage only, whereas deployment events are more closely associated with higher severity impacts with greater potential for injury). In further contrast, the FARS is specific to fatal collisions and primarily captures a subset of the highest severity deployment incidents and certain non-deployment or difficult-to-detect incidents involving certain object types.

Based on all these considerations, to calculate the major collision “U.S. average” estimate, Tesla divided the FHWA vehicle miles traveled (from all vehicles reported) by the estimated total count of vehicles involved in CISS-reported incidents (see the total value in Table 2 of the CISS reports).

The CISS and FHWA databases do not contain highway and non-highway breakdowns that are directly comparable to Tesla’s data. Therefore, to estimate a baseline for the “U.S. average” collision rate breakdown by highway and non-highway, Tesla computed the relative fraction of the overall collision rate seen in the manually driven Tesla fleet for both highway and non-highway driving. Tesla then applied those fractions to the major collision “U.S. average” rate to estimate the “U.S. average” collision rate for highway and non-highway driving. Tesla applied the same method to estimate the “U.S. average” rate for “Minor Collisions” by using the ratio of minor collisions to major collisions in the Tesla manually driven fleet. The Tesla manually driven fleet is a reasonably representative approximation for these fractions, due to its size (over three million U.S. vehicles), geographic distribution (presence in every U.S. state), owner makeup (Model Y was the second bestselling non-pickup in the U.S. in 2023-2024) and activity (over 30 billion miles annually in the U.S.).

Tesla appreciates that any adjustments to data can potentially introduce inadvertent noise and bias. To preserve the accuracy and integrity of the methodology used to calculate the U.S. average and compare it to Tesla collision rates, Tesla purposely limited any processing or filtering of data to be strictly minimal and only as necessary, as outlined above. Even still, the methodology involves necessary and unavoidable assumptions due to differences in data collection methods between Tesla’s data and publicly available data published by the U.S. government. These assumptions may contain limitations with respect to reporting criteria, unreported incident estimations (e.g., NHTSA estimates that 60% of property damage-only crashes and 32% of injury crashes are not reported to police [Blincoe et al. 2023]), federal database sample size and fleet distribution. Some of these limitations may skew the U.S. average calculation higher or lower than presented in the Vehicle Safety Report. Notwithstanding these limitations, the magnitude of improved real-world safety using FSD (Supervised) is clear and undeniable. This is plainly evident in the most direct comparison between Tesla vehicles using FSD (Supervised) and those being driven manually. The U.S. average estimate (even with its limitations) simply reinforces that conclusion.

Tesla collects collision telemetry to understand the nature of real-world collisions and develop solutions to prevent or mitigate them.

This web page presents data from different date ranges. Overall FSD (Supervised) mileage is presented from 2020 onward, which coincides with the launch of FSD (Supervised). The latest 12 months of collision rate metrics will be used each time the report is updated, which will be per quarter.

Tesla will update collision rate metrics in this report every quarter. The data will reflect a rolling twelve-month aggregation of miles and collisions in an effort to remain relevant to recent trends and progress.

Tesla's collision reporting methodology relies on automated telemetry data collected directly from vehicle systems. This approach enables objective collision detection across millions of vehicles. Other data, such as occupant injury information, is inconsistently provided through voluntary reporting by drivers or otherwise inaccessible to Tesla due to health-related privacy laws.

Instead, Tesla focuses on objective and programmatic metrics such as collision frequency and airbag deployment rates. Airbag deployments serve as a reliable proxy for collision severity. For example, airbag deployment thresholds are specifically engineered to activate airbags at impact severities where occupant injury risk increases significantly. Airbag non-deployment collisions typically represent lower severity collisions that may not require an airbag deployment. Together, these make for meaningful and objective safety indicators, which can be consistently measured across the entire fleet of Tesla vehicles. In addition to the regular quarterly updates, Tesla is committed to enhancing the Vehicle Safety Report with additional insights, statistics and reporting methodology.

Find information on Tesla vehicles’ crash safety.

NHTSA’s Standing General Order on Crash Reporting (SGO) requires manufacturers to report known collisions involving a vehicle traveling with an Automated Driving System (ADS) or Advanced Driver Assistance System (ADAS) engaged within the last 30 seconds before a collision occurs, among other reporting criteria. However, the SGO does not require manufacturers to develop a telemetry system to receive notification of collisions involving an ADAS or ADS system. Tesla uses the collision telemetry system described in this Vehicle Safety Report to supplement federal reporting obligations, including the SGO.

Tesla designed its vehicles to maximize connectivity and capability. We use remote telemetry to learn from the vehicle fleet to inform a variety of engineering decisions, including being notified about safety-critical events (such as major or minor collisions). In contrast, other vehicle manufacturers, whose vehicles are less connected, largely do not receive vehicle data when a collision occurs that involves one of their vehicles unless or until a customer tells them or sues them.

The consequence of the SGO’s reporting scheme suggests higher absolute collisions among Tesla vehicles than for every vehicle manufacturer combined; however, the reality is that Tesla reports more collisions simply because we have a large, active and fully connected customer vehicle fleet. This unfortunate consequence of the SGO is compounded by a lack of data normalization. While the SGO requires manufacturers to report absolute collision counts, it does not require manufacturers to report the number of their vehicles operating on U.S. roads or the distance traveled in those vehicles. If one considers absolute collision counts alone, collisions will appear to increase with increased adoption or usage, while in reality, the rate of incidents may be decreasing.

If Tesla vehicles are available for sale in your country, but FSD (Supervised) is not yet available, it is because your country may require explicit regulatory approval from local authorities. Tesla is actively working with governments worldwide to secure regulatory approvals, with successful rollouts already in place in the U.S., Canada, Mexico, Puerto Rico, Australia and New Zealand. While we can’t provide exact timelines for specific countries, progress is being made. Once FSD (Supervised) is approved for a given country, eligible vehicles will receive it with an over-the-air software update.