Tesla FSD Safety Stats Misleading

[snstevens]

https://electrek.co/2026/05/28/tesla-fsd-safety-stats-misleading-reuters-investigation/

Is Electrek reporting this Reuter’s investigation correctly?

Personally I find the idea of an entirely camera-based self-driving system to be dangerous since weather conditions anffect its reliability. However, I’ve never driven a Tesla with FSD so can’t comment. Perhaps others can add color.

Sponsored

 

[PorscheTaycan]

Personally I find the idea of an entirely camera-based self-driving system to be dangerous since weather conditions anffect its reliability.

Agree. Why Tesla wouldn't use tried and true tech like radar and even a rain-sensing module instead of cameras I don't understand...

 

[whitex]

Personally I find the idea of an entirely camera-based self-driving system to be dangerous since weather conditions anffect its reliability. However, I’ve never driven a Tesla with FSD so can’t comment. Perhaps others can add color.

Tesla is not the only one who's going there.

The top level argument is pretty simple - if humans can drive with 2 cameras and 2 microphones on a swivel plus a 3D accelerometer, so can AI (they add some cameras instead of putting them on swivel). The basic idea is that just like AI can write stories, code, other documents after being trained by reading billions of documents off the internet, or can create pictures and videos after being shown billions of pictures and videos, AI can also drive like humans after being shown recordings of billions of miles driven by humans.

The devil is of course in the details, as AI learns you need safety guiderails to keep its hallucinations from hard violating rules, etc. As AI improves, and gets more and more training data, it will only get better. Having radar, lidar and other sensors that humans don't have can be useful for the guiderails, but for training the AI it might actually be counterproductive since humans don't react to radar stimuli for example, so AI has nothing to learn from them.

Perhaps in the future we can have humans driving around with radar and lidar screens, then those recordings of all the cameras/lidars/radars can be used to train AI which can utilize all those sensors in its mimicking human driving. However, if AI can drive as good as best humans with just cameras, it makes no sense to make cars more expensive to equip them with additional sensors. I think V2V communications will be cheaper and much more effective as more and more traffic becomes self-driving (each car drives based on its own sensors, and sensors of all cars nearby).

 

[whitex]

Agree. Why Tesla wouldn't use tried and true tech like radar and even a rain-sensing module instead of cameras I don't understand...

Cost, and see my post above.

 

[PorscheTaycan]

if humans can drive with 2 cameras and 2 microphones on a swivel plus a 3D accelerometer, so can AI

I get the theory and believe that we will get there one day, even soon.

But as Tesla drivers know, humans can recognise with their eyes that it is raining, but the Tesla camera AI keeps getting it wrong, with dry wiping happening constantly, and the wipers not coming on when it's raining (something a rain sensor can detect with no issue in other cars).

 

[whitex]

I get the theory and believe that we will get there one day, even soon.

But as Tesla drivers know, humans can recognise with their eyes that it is raining, but the Tesla camera AI keeps getting it wrong, with dry wiping happening constantly, and the wipers not coming on when it's raining (something a rain sensor can detect with no issue in other cars).

I've had 4 Model S's over a decade. 2 of them had the camera based rain sensing. Yea, it sucked. However, I also experienced dry wiping with my Taycan just this past week (the only reason I even remember this is because there was some bird crap on the windshield which I meant to clean but was in a rush, figured I can do it after returning back home, but the dry wipe just spread the crap with the wiper so I regretted not taking the extra minute before leaving the house). I suspect if you get the latest Tesla with the latest AP4, the rain sensing will work ok - Tesla has had years to perfect it.

I took a Tesla RoboTaxi (I think it was a Model Y, not the 2 seater one) when I was in California a few weeks back. Yea, it had a safety driver, but it did really well navigating evening traffic in Silicon Valley, which included having to push through 5 lanes of traffic to turn right after entering a highway from the left. I was rather impressed, door to door, no safety driver interventions. Of course, the last 20% of the usecases take 80% of the effort, but still, it was interesting to see. The RoboTaxi was quite tame in its driving style by the way.

Others have shown impressive demos as well with world-model AI driving, Tesla is just pushing the hardest to get released to the public, taking a lot of risks too, but that's Elon's M.O.

 

[FlyingPoint]

In my own experience after having driven thousands of miles using FSD, it is not perfect. However, it is far safer in my experience than a human driving. It has gotten far better with with construction zones and emergency vehicles over the last 6 months. I never had an issue with motorcycles. If a motorcycle is lane splitting, then that's their problem not mine. Off ramps have been perfect.

The inclement weather issue has not been a problem. It has handled snow and heavy rain. If it determines that weather is impairing reliablility the driver is notified to take over.

The main point is that it continues to improve. Non statisically, it has improved 50% in the last year.

 

[4424]

In my own experience after having driven thousands of miles using FSD, it is not perfect. However, it is far safer in my experience than a human driving. It has gotten far better with with construction zones and emergency vehicles over the last 6 months. I never had an issue with motorcycles. If a motorcycle is lane splitting, then that's their problem not mine. Off ramps have been perfect.

The inclement weather issue has not been a problem. It has handled snow and heavy rain. If it determines that weather is impairing reliablility the driver is notified to take over.

The main point is that it continues to improve. Non statisically, it has improved 50% in the last year.

I agree completely, I have a 2023 Model X Plaid with Hardware 4 (the latest autopilot hardware) and it is much, much better than the average human driver. I can't remember the last time I had to take over because of a safety issue. I usually take over because I don't like the parking spot it is choosing or it is taking a route I don't like.

The last gen, HW3 is pretty good, but nothing close to HW4.

People who bash FSD have not been in a HW4 car with the latest software. It was awful about 18 months ago but has improved immensely. It is not perfect but much better than human drivers.

 

[Vercingetorix]

https://electrek.co/2026/03/17/former-uber-self-driving-chief-tesla-fsd-crash-supervision-problem/

I wholly believe self driving may be safer than the human population as a whole, it is not safer than human drivers who pay attention.

 

[69Mach390]

Tesla is not the only one who's going there.

The top level argument is pretty simple - if humans can drive with 2 cameras and 2 microphones on a swivel plus a 3D accelerometer, so can AI (they add some cameras instead of putting them on swivel). The basic idea is that just like AI can write stories, code, other documents after being trained by reading billions of documents off the internet, or can create pictures and videos after being shown billions of pictures and videos, AI can also drive like humans after being shown recordings of billions of miles driven by humans.

The devil is of course in the details, as AI learns you need safety guiderails to keep its hallucinations from hard violating rules, etc. As AI improves, and gets more and more training data, it will only get better. Having radar, lidar and other sensors that humans don't have can be useful for the guiderails, but for training the AI it might actually be counterproductive since humans don't react to radar stimuli for example, so AI has nothing to learn from them.

Perhaps in the future we can have humans driving around with radar and lidar screens, then those recordings of all the cameras/lidars/radars can be used to train AI which can utilize all those sensors in its mimicking human driving. However, if AI can drive as good as best humans with just cameras, it makes no sense to make cars more expensive to equip them with additional sensors. I think V2V communications will be cheaper and much more effective as more and more traffic becomes self-driving (each car drives based on its own sensors, and sensors of all cars nearby).

Isn’t the point to be “better than humans?”

 

[FlyingPoint]

Look at this accident I witnessed a week ago. There is no doubt in my mind that if the victim car had FSD, it would have been avoided. The Tesla would have seen the car coming over in the right rear and moved over.

 

[snstevens]

https://electrek.co/2026/03/17/former-uber-self-driving-chief-tesla-fsd-crash-supervision-problem/

I wholly believe self driving may be safer than the human population as a whole, it is not safer than human drivers who pay attention.

Perhaps the relevant question from this is whether we can ever get past the supervision question. I found these two paragraphs from the @Vercingetorix referenced article chilling.

“What makes Krikorian’s account so compelling is that he’s not some random Tesla critic. He built self-driving cars at Uber. He trained safety drivers on intervention protocols. He understood the risk intellectually, and he still got conditioned into complacency. If someone with that level of expertise can get caught, the average Tesla owner doesn’t stand a chance.

“The “supervised” label is a legal shield, not a safety solution. Tesla knows that humans cannot reliably supervise a system that works 99% of the time — the research is clear, and Krikorian lays it out plainly. Yet the company continues to sell “Full Self-Driving” while pointing to the fine print when things go wrong.”

 

[snstevens]

Look at this accident I witnessed a week ago. There is no doubt in my mind that if the victim car had FSD, it would have been avoided. The Tesla would have seen the car coming over in the right rear and moved over.

Yikes!!

I see you pulled over to help. Was anyone hurt? Also, what did the driver say who caused the accident?

 

[snstevens]

https://electrek.co/2026/05/28/tesla-fsd-safety-stats-misleading-reuters-investigation/

Is Electrek reporting this Reuter’s investigation correctly?

Personally I find the idea of an entirely camera-based self-driving system to be dangerous since weather conditions anffect its reliability. However, I’ve never driven a Tesla with FSD so can’t comment. Perhaps others can add color.

Conversation so far has been interesting, but no one has commented on the point of the Reuter’s article that says the Tesla employees who train FSD don’t trust it.

I’m having trouble getting past that.

 

[FlyingPoint]

Yikes!!

I see you pulled over to help. Was anyone hurt? Also, what did the driver say who caused the accident?

Occupants were banged up but nothing too serious. Luckily there was no fire. The other driver didn't speak english, so not much conversation with him. I must say it was a bit unsettling, however my 15 years of track experience prepared me for exactly what to do in case the car shot back across the highway. Had this occurred 200 yards earlier or later, the guardrails would have created a much different ending.
As I look at the video, I just don't get what would cause someone to switch lanes so aggressively. These cars had no interaction previously, so no road rage.

Sponsored