Tesla Says Autopilot Is Getting Better. But It

• Tesla’s new Autopilot safety report shows crashes rising for a third straight quarter.
• Tesla claims Autopilot users are nine times less likely to crash than unassisted drivers, but that’s down from ten times safer earlier this year.
• Tesla still withholds detailed Autopilot safety data, so we can’t really have a clear picture of how safe it really is.

Tesla has never released a detailed safety report on Autopilot or Full Self-Driving (FSD), probably because it might paint the automaker-turned-autonomous-vehicle-and-artificial-intelligence company in a bad light. What it does share is a single statistic each quarter: how often drivers on Autopilot crash compared to how often U.S. drivers crash overall. But despite this highly favorable statistic, the company’s most recent reports have actually shown a decrease in safety, and the latest one, released for Q3 2025, continues on a downward trend.

The headline sounds pretty good: Tesla says drivers who use Autopilot are around nine times less likely to crash compared to the U.S. national average. The manufacturer notes that its latest safety report shows “significant advancements in its Autopilot technology,” even though what it actually shows is falling numbers for a third consecutive quarter.

According to the new report, vehicles equipped with Autopilot experienced one crash for every 6.36 million miles driven. Tesla points to 2023 data from the National Highway Traffic Safety Administration (NHTSA) and the Federal Highway Administration (FHWA), which states that the national average of miles driven before a crash is around 702,000 miles.

Big picture, the numbers have improved considerably since Q3 2018, when it filed the first such report. Back then, drivers who used Autopilot had one crash every 3.35 million miles, while those who didn’t use it drove 1.92 million miles before experiencing a crash and the national average was 481,000 miles.

However, looking at Tesla’s Q1 2025 safety report, it shows that Autopilot users drove 7.44 million miles before a crash, while those who didn’t use it crashed, on average, after 1.51 million miles. Autopilot’s peak safety quarter so far has been Q1 2024, with 7.63 million miles between crashes with Autopilot on and 955,000 miles with it off.

I would love to dig through more detailed Autopilot and FSD safety data. How many times does it disengage? How many times did drives have to take over to avoid a potential hazard? How many of these Autopilot miles were driven on the highway, and how many in town? These are just some questions that we can’t answer because—while Tesla likely has all of it in granular detail—it won’t release it.

Tesla presents Autopilot as a way for drivers to enhance safety and reduce the workload, which it is. The system works well enough and helps take some of the workload out of driving. It is not fully self driving, however, and neither is Tesla’s Full Self Driving package, despite the misleading name. Autopilot relies on Autosteer and what Tesla calls “Traffic-Aware Cruise Control,” which basically means it adjusts its speed depending on what’s in front of the car and around it and it steers to keep you in your lane.

More On This

It really does take the strain out of driving, especially on long journeys, and that’s the key to putting Tesla’s safety report numbers into context. Most people will likely choose to use Autopilot for highway driving. Since there are no intersections with crosswalks, no bike lanes or other pieces of urban road infrastructure and little chance of pedestrians accidentally stumbling onto the road from behind a parked car, driving on the highway is generally safer and less eventful. That’s why Tesla recommends only using Autopilot on closed highways, while FSD is designed to work everywhere. 

So all those miles driven with Autopilot engaged, where it’s that much safer than the average for human drivers, is only part of the story. The national statistic includes driving on all types of roads, including in towns where the chance of having a crash is higher. And we also don’t know what qualifies as a crash for Tesla. Does a low-speed bump in city traffic count?

It will be interesting to see what future safety reports will show as people start to buy the new Standard versions of the Model 3 and Model Y, which don’t get Autopilot as standard. They will still maintain a safe distance to the car in front, but won’t keep the car centered in its lane if owners don’t pay $8,000 for the FSD package, which they probably won’t if they’re trying to save a buck by buying the cheapest available models. But the real test for Tesla’s autonomous ambitions is whether it can safely remove drivers altogether with its robotaxi service, which has begun its slow roll-out. Musk says new cities are coming this year, so we should learn more soon.