When the Computer Was Driving
Who Is Responsible?

The crash report says the driver was on Autopilot. Or the vehicle was a Waymo robotaxi with no human at the wheel. Or the car was on GM Super Cruise on a highway segment the system was supposed to handle and instead drifted into a stopped emergency vehicle. The body in the hospital bed is real. The question of who is on the hook is genuinely unsettled in American law, and the answer changes depending on what level of automation was engaged, what the manufacturer told the public the system could do, and what the data inside the vehicle shows about the seconds before impact.

This is a primer on how a plaintiff team approaches an automated-driving crash case.

The SAE levels, plainly

The Society of Automotive Engineers (SAE) Levels 0–5 are the framework everyone in the field uses to describe what a vehicle's driving system actually does. The federal government uses them. The manufacturers cite them in their disclaimers. Juries are introduced to them at trial. In plain English:

• Level 0: The human does everything. Warnings only, like blind-spot indicators.
• Level 1: One function automated — e.g., adaptive cruise control or lane-keep assist. Human still in charge.
• Level 2 (most "self-driving" features sold today): The system can steer and control speed simultaneously under some conditions. The human is required to monitor and is responsible at all times. Tesla Autopilot, Tesla Full Self-Driving, GM Super Cruise, Ford BlueCruise, and most other consumer systems live here.
• Level 3: Under defined conditions, the system handles the driving and the human can stop monitoring — but must be ready to take over on request. Mercedes Drive Pilot is the only Level 3 system approved on U.S. public roads as of this writing, and only in limited operating conditions.
• Level 4: The system handles the driving entirely within a defined operating domain — specific cities, weather, time of day. No human driver needed inside that domain. Waymo's commercial robotaxi service is the leading U.S. example.
• Level 5: Full automation everywhere a human could drive. No vehicle on a U.S. public road is at Level 5.

The single most important fact a plaintiff team has to establish, early, is the level of the system involved in the crash and whether it was actually engaged at the moment of impact.

Why "Autopilot" and "Full Self-Driving" are a special category

Tesla's marketing language for its Level 2 systems — "Autopilot" and "Full Self-Driving" — has been the subject of more product-liability scrutiny than any other automated-driving consumer feature. NHTSA's Office of Defects Investigation has opened multiple investigations into Autopilot performance, including a series of crashes involving stationary emergency vehicles. The agency's 2023 recall covering more than 2 million Tesla vehicles required the company to update its driver-monitoring system, in part because the public-facing system name had outpaced what the technology could actually do.

The legal theory the plaintiff bar has developed against Tesla has three parts. First, the system was marketed in a way that foreseeably caused drivers to over-trust it. Second, the in-vehicle driver-monitoring was inadequate to keep distracted or inattentive drivers from misusing the system. Third, the system has documented performance failures — phantom braking, failure to recognize stopped vehicles in the travel lane, failure to recognize cross-traffic in some scenarios — that were known to the company and not adequately disclosed or fixed. A 2023 California state court verdict and several federal cases have tested these theories with varying outcomes.

None of this means the human driver is off the hook. Tesla's terms of use and the on-screen warnings explicitly tell the driver to monitor at all times. A typical case names the driver and the manufacturer in the alternative, and the trier of fact apportions responsibility.

Robotaxi cases: when there is no human driver

A driverless Waymo or a driverless Cruise vehicle (during the period before Cruise's California suspension) presents a different problem. There is no human driver. The only entity that controlled the vehicle is the automated driving system, owned and operated by the company. The case proceeds as a traditional negligence claim against the operator, paired with a product liability claim against the manufacturer (often the same corporate family). The defenses are no longer "the driver wasn't paying attention." The defenses become "the system performed reasonably under the circumstances" or "the third party caused the crash."

Robotaxi cases are evidence-rich and evidence-controlled by the defendant. Sensor logs, internal incident reports, prior near-miss data for the same intersection or scenario, and the operator's post-incident investigation are all subject to civil discovery. The cases that get the deepest into the company's data are the ones that ultimately move the needle.

The data inside the vehicle

Modern automated and semi-automated vehicles record a lot. The categories that matter to the case:

• System engagement logs: exactly when the Level 2 system was on, when it disengaged, and whether the disengagement was driver-initiated or system-initiated. Tesla's data has shown system disengagements in the final seconds before crashes, which the company has occasionally cited to argue the driver was in control at impact.
• Sensor data: what the cameras, radar, and (for some vehicles) lidar were seeing. Whether the system identified the hazard at all.
• Driver monitoring data: hands-on-wheel torque, eye-tracking camera data on systems that have it, alert history.
• Event data recorder (EDR): the same "black box" data available in any modern vehicle — speed, brake application, throttle position, steering input, seatbelt status, airbag deployment. Our companion guide on what an event data recorder shows covers that piece.
• Over-the-air update history: what software version was running, when it was installed, and what the release notes said.

The evidence is on a server controlled by the manufacturer. A preservation letter goes out within days, and a forensic protocol is set up before the data is downloaded. The first 24 hours after the crash matter more in these cases than in any other vehicle case — see the first 24 hours.

The NHTSA Standing General Order

In 2021 the National Highway Traffic Safety Administration issued a Standing General Order requiring manufacturers and operators of Level 2 driver-assistance systems and Level 3–5 automated driving systems to report crashes to the agency. The reports include vehicle identification, the system involved, basic crash circumstances, injury severity, and follow-up data. The public summaries are searchable and constitute one of the most useful pieces of public evidence available in any automated-driving case.

Plaintiff teams routinely cross-reference the Standing General Order data with their own incident, looking for the same system, the same software version, and the same failure mode in other crashes. Pattern evidence of repeated failures across crashes is some of the most persuasive material a jury sees.

Three liability tracks, sometimes all in one case

An automated-driving crash case typically pursues some combination of:

• Negligence against the human driver (Level 2 systems). The driver was responsible for monitoring and did not.
• Product liability against the manufacturer. Design defect (the system was unreasonably dangerous as designed), manufacturing defect (a defect in the specific vehicle), or failure to warn (the system's limitations were not adequately disclosed).
• Negligence against the operator (robotaxi cases). The system's behavior fell below the standard of care for a reasonable automated driving system in the same conditions.

The right combination depends on the facts, the jurisdiction, and which defendant has the resources to make the case meaningful. Sorting that out is the early-case work.

What this means for a case

If you or someone you love was hit by a vehicle that was being driven by a partly- or fully-automated system, three things matter. First, preserve the evidence inside the vehicle that hit you — a litigation hold letter to the manufacturer and the operator goes out fast. Second, do not let any insurance carrier settle this case as a routine motor-vehicle claim — the product liability exposure changes the value picture significantly. Third, get the case in front of an attorney who has actually litigated against the data discovery these companies will throw at the case. The technology is new. The injuries are the same as in any catastrophic crash. The path to accountability is genuinely different.