This as-told-to essay is based on three conversations with Brett Schreiber, a California-based lawyer who represented the plaintiffs in the Miami Autopilot case involving Naibel Benavides Leon. In 2025, a federal jury found Tesla partially responsible for the crash that killed Benavides Leon and seriously injured Dillon Angulo. The jury ordered Tesla to pay more than $242.5 million in damages. Tesla is appealing the verdict. Tesla and its attorney did not respond to requests for comment. The conversations have been edited for length and clarity.
I’ve practiced law in California since 2005.
I came up in a plaintiff civil litigation firm in San Diego and had the opportunity to work up and try serious-injury cases.
Around 2021, I started getting calls about Tesla crashes. At the time, I had a GED-level education in autonomous vehicles, but that already put me head and shoulders above many of my contemporaries.
Now, my firm has weekly meetings called “Tesla Autopilot Review Team” and gets dozens of calls a month from across the country on incidents involving Autopilot or Full Self-Driving modes.
The vast majority of them involve minor or modest injuries that I’m simply incapable of reviewing. The barrier to entry due to the cost of litigation is so high.
In the Miami case, I spent over a million dollars on litigation costs alone — not including fees, but experts, depositions, data work, and everything else. I’m a contingency lawyer. The math on that has to be a $10-plus million case to make sense.
So we have to limit the cases that we look at to those involving catastrophic injuries or, unfortunately, a loss of life.
For someone who is minimally hurt: great for them, bad for their lawsuit. At the end of the day, the courthouse doors are closed to most of those people.
The Benavides case
The Benavides case was always about shared responsibility. We never tried to frame it as anything but that.
In every one of these Tesla Autopilot cases, there is typically an at-fault driver. Someone did something wrong. Someone failed to be an alert and attentive driver. That, however, should not absolve Tesla.
Tesla’s frame was to focus on the last three seconds: What was the driver doing? What didn’t they do? What systems were on? What systems were off?
Brett Schreiber
My role was to make it about something larger. Accidents happen, but systemic failures are caused by choices. If I made this case about the last three seconds, I would lose.
We can’t change the facts about the driver’s conduct. We talked about it. But we also argued that Tesla made decisions years before the collision — decisions to overstate, overhype, and oversell this technology, resulting in people over-relying on it just as this driver did.
So suddenly, it’s not just about the driver’s actions. What this driver did became inevitable, and that inevitability exists because of the choices Tesla made.
There is no playbook for AV lawsuits
In the world of auto-product liability, such as a tire-failure case or a seatback-failure case, there’s a bit of a playbook. Meaning, a lot of the same experts are involved. Testing has been done.
In autonomous vehicle technology, up until last year, there was no playbook.
The Benavides trial in Miami was only the third case ever to go to trial against Tesla, and the first time the plaintiffs won. We’re in the very nascent stages of this type of litigation. We’re building the plane as we’re flying it.
In the last half-century of auto product-defect litigation, you could typically point to a mechanical failure mechanism. Some bolt failed, some hose disconnected, or a seatback bolt wasn’t properly torqued, and as a result, the seat folded like a beach chair.
Here, if there are defect exists, they exist in the software. There is typically no mechanical thumbprint, no witness mark that shows you how the system failed. All of this is algorithmic. All of this lives in a world of ones and zeros.
The challenge is to turn algorithmic decision-making into something that we can discover and explain to jurors so they can understand how and why these crashes occurred.
Sometimes it’s a camera-fusion failure. Sometimes it’s a failure of automatic emergency braking or forward-collision warning to work. Sometimes a system activates in a place it should not have.
I don’t need Tesla’s source code to understand when a camera-fusion failure occurs. If three cameras target an issue and all three read it differently, I can show a jury that none of them could agree and there was no decision-maker.
The updates leave a glaring problem
There are a lot of Tesla vehicles still on the road today that use cameras and radar for the sensors.
People purchased these vehicles over the last decade with the idea or promise that they would reach a point where they would be fully self-driving and fully autonomous.
Elon Musk only recently said that not all Teslas will operate fully autonomously.
[In April, Musk said older Teslas with the Hardware 3 computer could not achieve unsupervised FSD without a hardware upgrade.]
Tesla has tried to push out over-the-air updates. We’ve seen in a number of areas that that’s kind of like trying to push out an update for an iPhone 17 and make it fit on an iPhone 4. There are computing limitations.
Flash forward: We’re now starting to see problems with the camera-only system.
[There is a current National Highway Traffic Safety Administration investigation into Tesla’s “degradation detection system.” A filing from NHTSA’s Office of Defects Investigation shows that the department is examining whether the system properly detects when conditions such as glare degrade camera visibility and warns drivers in time to respond.]
I believe in autonomy
I still fundamentally believe that autonomous driving is a tremendous societal good.
I believe that it is where the future of driving technology is headed, and it can be done well. This is not an indictment of the autonomous vehicle industry.
I think Tesla has to do two things.
One, the company has to be more intentional about the safety culture and about fixing bugs and problems far sooner.
The problem with an over-the-air software update is that it’s one thing if I haven’t updated my Spotify app. No one’s going to lose their life, whether I’m running the most up-to-date version of Spotify. But someone can lose their life if they’re not running the most up-to-date version of an autonomous vehicle software.
Number two, Tesla has to change the narrative that this is some glorious, full self-driving vehicle when it’s not.
There is only one company in the world that markets partial automation as “full self-driving.” The words partial and full in any reasonable interpretation of the English language do not mean the same thing.
I’m sorry, but calling it “Full Self-Driving Supervised” is not enough.
Have a tip? Contact this reporter via email at [email protected] or Signal at lloydlee.71. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.
