December 29, 2019. Gardena, California.
A Tesla Model S left the freeway, ran a red light, and hit a Honda Civic at an intersection. Gilberto Alcazar Lopez and Maria Guadalupe Nieves-Lopez died at the scene. Behind the wheel was Kevin George Aziz Riad, a limousine-service driver.
Three and a half years later, Riad pleaded no contest in Los Angeles County Superior Court to two counts of vehicular manslaughter with gross negligence, and was sentenced, that same day in June 2023, to probation instead of the more-than-seven-year prison term he had faced. The case is widely described as the first felony prosecution in the United States of a driver in a fatal crash where Autopilot was engaged.
That sentence does the easy work. The harder question, the one litigators should actually sit with, is what turned this from “a driver ran a red light” into “the first felony Autopilot prosecution” in the first place. The answer is not the brand name on the dashboard. It’s the data.
The gap the case turned on
“Autopilot” is a marketing word. California’s DMV alleged in 2022, and a state administrative-law judge found in December 2025, that Tesla’s “Autopilot” and “Full Self-Driving” branding was misleading under state law. But a criminal courtroom doesn’t work off a brand name. It works off the vehicle’s own record of what the system was doing, second by second, before impact.
Here, that record was decisive. NHTSA confirmed that the car’s Autosteer and Traffic-Aware Cruise Control were engaged at the time of the crash. Tesla’s own vehicle data, introduced through an engineer’s testimony at the preliminary hearing, showed the driver’s hand on the wheel and no braking for roughly six minutes before the collision. That’s what separates “a driver ran a red light” from a felony manslaughter case: the log, not the label on the dashboard.
That’s the artifact. The vehicle’s event-data record: Autopilot engagement state, steering and brake telemetry, speed, and whatever driver-monitoring signal existed. Not the word “Autopilot” on a spec sheet, but the log the vehicle actually kept.
Why a generalist misses it
A lawyer reading a Tesla brochure can tell you what Autopilot is supposed to do. A generalist engineer can tell you, in general terms, how driver-assist systems work. Neither of those things wins or loses a vehicular manslaughter case.
What decides it is narrower. What time resolution does the log actually have: can it place an input to the second, or only to the nearest several seconds? What did the system warn, and when? What did the driver do with the wheel and the pedals across those six minutes? Was this a failure of the system, a failure of the driver to respond to it, or both? And can the data actually tell those apart, or is someone reading a conclusion into a file that cannot carry it?
A court-appointed software expert doesn’t start with an opinion. They start by naming the materials: the event-data record or equivalent log, the Autopilot state history, the steering and brake telemetry, and the crash reconstruction built from the physical scene. Then a method: how each artifact was extracted, what its true time resolution is, where it can be corrupted or incomplete, and how the pieces cross-check against one another. Then, and only then, a conclusion, stated with its limits attached.
That last part is what survives cross-examination. An expert who says “the data shows exactly this, and here is what it cannot tell you” is credible. An expert who oversells a log file is not, and a good cross-examiner will find the seam.
Why Israeli counsel and tech GCs should care
Israeli automotive-tech, ADAS, and mobility companies sell systems with names like “Autopilot,” “co-pilot,” “assist,” “autonomous.” Those names carry legal weight the moment a customer, a regulator, or an injured party relies on them, and as the California ruling shows, the name itself can become the regulator’s target. This case is a preview of the question every one of those companies will eventually face: does the product’s telemetry actually show the system behaved the way the marketing said it would?
For litigators, the practical move is the same one that applies to any software-driven dispute: don’t wait for the crash-reconstruction expert to arrive with conclusions already formed. Ask, early, what data the vehicle actually kept, how it can be extracted intact, and who can testify to chain of custody. For GCs at companies building driver-assist or autonomous features, the same question runs the other way: can your own event logs support the claims your marketing team makes, under cross-examination, years after the incident?
The takeaway
The Riad case will keep getting cited as “the first felony Autopilot case.” That’s the easy headline. The harder lesson is that it turned on a narrow technical record, not a brand name: what the system did, what the driver did, and whether the vehicle’s own log answers that with enough precision to hold up in court. That’s not a question a generalist answers convincingly. It’s a court-grade software expert’s job.
The above is general information only and does not constitute legal advice. Case facts are drawn from the sources cited.