Tesla Autopilot, a Fatal Crash, and the Disappearing Data

On the night of April 25, 2019, a Tesla Model S running on Autopilot plowed into a parked Chevy Tahoe on a darkened two-lane road in Key Largo, Florida. The driver, George McGee — a private equity managing partner who had bought the Tesla specifically to use Autopilot on his 100-mile night commutes — had dropped his phone and bent down to retrieve it. When he looked up, it was too late. Dillon Angulo survived with a shattered pelvis, broken jaw, and bleeding brain. His girlfriend, 22-year-old Naibel Benavides, was thrown over 115 feet and died. They had been standing outside the car, looking at the stars.

The crash quickly became something far larger than a local traffic homicide. Florida Highway Patrol Investigator Cpl. David Riso launched a criminal investigation and moved to obtain the most critical piece of evidence: the Autopilot computer, a flat silver box tucked above the glovebox, which should have stored five seconds of pre-crash video, real-time object detection data, and a complete picture of what the system perceived — and failed to do — in its final moments. Every Tesla with Autopilot is engineered to automatically transmit crash data to company headquarters. The data, however, is stored in a proprietary format that only Tesla can decode, making investigators entirely dependent on the company's cooperation.

What followed was a years-long battle over that data. Riso was directed to a Tesla service technician named Michael Calafell, who swapped the crash computer into a similar vehicle at a Coral Gables service center. About 30 minutes later, Calafell delivered the verdict: the file was corrupted. Nothing of value. Tesla's deputy general counsel Ryan McCarthy — a former product liability defense attorney who had spent 13 years defending Tesla before joining the company — subsequently provided Riso with just two items: infotainment logs showing McGee's 13-minute call with American Airlines (which ended two seconds after the crash), and forward-facing dashcam footage of McGee running the stop sign. The Autopilot snapshot — the data showing exactly what the system identified and failed to warn — was gone. Riso closed the case over a year later.

The missing data sits within a broader pattern of regulatory frustration with Tesla. The NTSB had warned the company after a 2016 Florida fatality involving Navy veteran Joshua Brown to restrict Autopilot to divided interstate highways with no intersections or crossing traffic. Tesla never did. After Apple engineer Walter Huang died in 2018 when his Model X slammed into a highway divider — with Tesla initially blaming him for playing games on his phone — the company settled the lawsuit the day before trial. In early 2019, Jeremy Banner died when his Autopilot-equipped Model 3 struck a tractor-trailer at 68 mph, his car continuing to drive his body down the highway for 40 seconds afterward. The NTSB's chairwoman later wrote directly to Elon Musk noting Tesla had not responded to the agency's safety recommendations. NTSB Vice Chair Bruce Landsberg called Tesla's Autosteer 'completely inadequate' in 2020.

After Naibel Benavides's family sued Tesla in April 2021, the case drew in independent safety researchers and a reclusive Russian-born hacker known only as Green, who became central to efforts to recover or reconstruct what the Autopilot computer had seen. The case evolved into one of the most consequential legal confrontations in the history of autonomous vehicle technology — a referendum not just on whether the system failed that night on Card Sound Road, but on whether Tesla systemically controls, withholds, or loses the evidence that could hold it accountable when it does.