Autonomous Vehicle Liability — Who's Responsible When the Car Crashes Itself?

---
title: Autonomous Vehicle Liability — Who's Responsible When the Car Crashes Itself?
slug: autonomous-vehicle-liability-framework
tags: autonomous-vehicle,liability,regulation,av
---

In October 2023, a Cruise robotaxi in San Francisco struck a pedestrian who had been thrown into its path by another vehicle. The Cruise vehicle braked as designed — but then pulled forward, dragging the injured pedestrian 20 feet. The incident led to the suspension of Cruise's operating license in California. It also crystallized a legal question that the automotive and insurance industries have been circling for years without answering: when an autonomous vehicle causes harm, who is responsible?

The answer, it turns out, depends heavily on what level of automation is involved — and the SAE automation levels that engineers use as a technical classification have been quietly becoming the liability assignment framework that courts and regulators are building their rules around.

## The SAE Level Framework as Liability Architecture

The SAE International J3016 standard defines six levels of driving automation. The liability implications diverge dramatically across these levels.

**Level 2 (Partial Automation)** — Tesla's Full Self-Driving Supervised, Ford's BlueCruise, GM's Super Cruise — keeps the human driver as the legally responsible party. The driver must remain attentive and ready to take over at any moment. The system provides steering and speed control as driver assistance, not as autonomous operation. Every NHTSA investigation of a Tesla FSD-related crash has assessed driver responsibility under this framework.

**Level 3 (Conditional Automation)** introduces the first significant liability transfer to the manufacturer. The driver may disengage from the driving task within the system's operational design domain — hands off the wheel, eyes off the road — but must be ready to resume control when requested. Honda's Legend, which received Level 3 type approval in Japan in 2021, and Mercedes-Benz's Drive Pilot, approved for Level 3 operation in Germany and in Nevada, explicitly accept manufacturer liability for incidents occurring during Level 3 engaged operation within defined conditions (low-speed highway, clear weather, specific geographic boundaries).

**Level 4 and 5 (High/Full Automation)** — Waymo's commercial service, Cruise's paused service, Zoox's operations — shift liability away from the vehicle occupant entirely. There is no driver to be liable. The operator (the company running the service) and the manufacturer bear responsibility for incidents.

---

## The Product Liability Standard

The transition from driver liability to manufacturer liability is not merely a matter of insurance policy — it represents a fundamental shift in the legal theory under which AV incidents are adjudicated.

Driver negligence cases operate under a *negligence standard*: did the driver fail to exercise reasonable care? Manufacturer liability for AV incidents operates under *product liability* doctrine: did the manufacturer design, manufacture, or provide adequate warnings for a product that was unreasonably dangerous?

Product liability in the US operates under three distinct theories. **Manufacturing defect** — the specific vehicle deviated from its intended design. **Design defect** — the entire product line has a design that makes it unreasonably dangerous. **Failure to warn** — the manufacturer did not adequately inform users about known risks or limitations.

For AVs, design defect is the most consequential theory. If a manufacturer's software makes a decision that causes harm — choosing to brake sharply when a slower deceleration would have been safer, misjudging the speed of an approaching vehicle, failing to detect a pedestrian in unusual lighting conditions — the plaintiff's claim is that the software design itself was defective. This is a product liability claim, not a negligence claim, and the standard of proof and the potential damages exposure are substantially different.

## Arizona, California, and the Regulatory Patchwork

In the absence of comprehensive federal AV legislation, the regulatory framework for autonomous vehicle operations has been assembled state by state, creating a patchwork that reflects each state's economic interests and risk tolerance.

**Arizona** became the de facto home for AV development not by accident but by deliberate policy. Governor Ducey's 2015 executive order welcomed AV testing without requiring state permits, and subsequent legislation created one of the most permissive regulatory environments in the US. Waymo launched its commercial robotaxi service — with no safety driver — in the Phoenix suburbs in 2020, years before similar operations were commercially viable in California or any other state.

**California** operates the most complex regulatory framework. The California DMV issues Autonomous Vehicle Testing Permits and, separately, Deployment Permits for commercial operations. Following the Cruise incident, the DMV suspended Cruise's permit and conducted a review that revealed Cruise had provided incomplete information about the pedestrian dragging incident — the data monitoring and reporting obligations that now attach to AV operators in California are substantially more rigorous as a result.

**Federal authority** rests primarily with NHTSA, which regulates Federal Motor Vehicle Safety Standards. NHTSA's Standing General Order on Crash Reporting (SGPO) requires manufacturers and operators to report crashes involving Level 2+ automation. The data collected under SGPO — covering hundreds of thousands of AV-involved incidents — is the primary empirical basis for understanding how AV systems actually perform in the field.

## The Insurance Industry's Transition Problem

Personal auto insurance is priced and structured around a model where the human driver is the primary risk factor. Driver age, history, driving patterns, and behavior drive actuarial modeling. In an AV context — particularly Level 3 and above — the human behavior variable either disappears or changes character.

Several insurers have begun offering AV-specific coverage products, but the actuarial challenge is substantial: the data to price Level 4+ AV risk accurately doesn't yet exist at scale because the operational fleets are too small and too new. The frequency and severity distributions for AV incidents may differ dramatically from human-driver incidents in ways that won't be visible for years.

The likely long-term structure — already visible in Level 3 frameworks like Mercedes Drive Pilot — is a hybrid model where manufacturer-provided coverage handles incidents during autonomous operation and personal auto coverage handles incidents during human-controlled operation. The handoff between these regimes, and how it's adjudicated in edge cases, is an active area of both regulatory and industry development.

## The Black Box Problem

One lesson the aviation industry learned from decades of incident investigation is that outcome analysis requires comprehensive data recording. Commercial aircraft have mandatory flight data recorders and cockpit voice recorders precisely because reliable post-incident analysis is impossible without them.

AV regulatory frameworks are converging on a similar requirement. NHTSA's proposed AV safety reporting requirements, California's DMV reporting obligations, and equivalent frameworks in Germany and Japan all create data collection and preservation requirements designed to ensure that post-incident investigation can reconstruct what the AV system knew, what decision it made, and why.

The data asymmetry problem remains: manufacturers possess detailed telemetry from their vehicles; plaintiffs and regulators typically access only what manufacturers are required to disclose. The scope of mandatory disclosure requirements will substantially shape whether the liability framework actually produces appropriate accountability — or whether manufacturers can selectively present data in ways that obscure the relevant decision sequences.

## The Verdict on Where This Goes

The liability framework for autonomous vehicles is being built in real time, through regulation, litigation, and the accumulation of incident precedent. The SAE level structure has become the de facto organizing principle. The direction — toward manufacturer and operator liability as automation levels increase — is clear. The details of how claims are valued, how data is accessed, and how international jurisdictions align their frameworks are still being worked out.

One number tells much of the story: Waymo's commercial fleet, as of early 2026, has driven tens of millions of miles in complex urban environments with safety performance that compares favorably to human baseline statistics. The technology may be solving the safety problem even as the legal framework struggles to catch up with it.

Autonomous Vehicle Liability — Who's Responsible When the Car Crashes Itself?

// COMMENTS

ON THIS PAGE