Autonomous Vehicle Liability: Who Is Responsible?

Autonomous vehicle liability is the question of who bears legal responsibility when a vehicle controlled partly or wholly by software is involved in a crash. There are three honest answers, and which one applies depends on the automation level and the jurisdiction.

In a Level 2 car, the one almost certainly in your driveway, the answer is settled and uncomfortable: you are. NHTSA states plainly that “Every vehicle currently for sale in the United States requires the full attention of the driver at all times for safe operation.” In the United Kingdom, Parliament wrote a different answer into law for genuinely self-driving vehicles. In the United States, for driverless fleets, the answer is still being assembled case by case and state by state.

Key Takeaways

• Level 2 is not ambiguous. When a car steers and holds speed but requires supervision, the human driver remains responsible for the driving task. Marketing names do not change this.
• The UK’s Automated Vehicles Act 2024 created an “authorised self-driving entity” in section 6 and, in section 47, a heading that resolves the central question: “User-in-charge not liable for manner of driving.”
• The United States has no equivalent federal statute. The National Conference of State Legislatures records 29 states and Washington D.C. having enacted autonomous vehicle legislation, with governors in 11 more states acting by executive order, and some states setting liability coverage minimums as high as $2,000,000.
• In practice the active levers today are product liability, insurance, federal crash reporting and recalls rather than a purpose-built liability regime. Rivian’s October 2, 2026 R2 recall over poorly tightened battery packs is an example of the recall mechanism doing the work.
• An investigation is not a finding. NHTSA opened an investigation into Comma’s hands-off driving technology on September 23, 2026 after two fatal crashes. No cause has been determined and no defect has been found.

Why the automation level decides the answer

Liability tracks control. If a human is required to monitor and intervene, the law treats that human as the driver. If the system is authorised to drive without supervision, someone else must answer for it, because there is nobody in the seat to blame.

That is why the SAE level is not a technical curiosity but the pivot of the entire legal question. And it is why the vagueness of the marketing vocabulary matters so much. As the reference summary of the field puts it, “As of 2026, the term ‘self-driving’ lacks an agreed standard definition and is also subject to commercial advertising and branding considerations.” A consumer who believes they bought a self-driving car and in fact bought a Level 2 driver-assistance system has misread their own legal position, not just a spec sheet.

The Insurance Institute for Highway Safety sharpened this point from the data side. Its research found “no crash-reduction advantage for vehicles equipped with partial driving automation compared with vehicles from the same automakers that were only equipped with crash avoidance technologies,” and it observes that “drivers often have a false sense of security about what the technology can do.” A false sense of security is a liability problem wearing a psychology costume. If you are supervising a system you believe does not need supervising, you are the responsible party and you are not performing the role.

One more data point for perspective on how unsettled this all is: Mercedes-Benz offered a Level 3 traffic-jam feature up to 40 mph until January 2026, then removed it in favor of an upcoming Level 2 capability. The one consumer system that had formally crossed the line where the manufacturer accepts the driving task was pulled back.

The United Kingdom wrote the answer down

The Automated Vehicles Act 2024, 2024 c. 10, is the most legible attempt by any major jurisdiction to answer this question in statute rather than in litigation. Three pieces matter for a parent trying to understand the shape of it.

Section 6 creates the authorised self-driving entity. Before a vehicle can be operated as self-driving, an identified legal entity must be authorised and must stand behind the vehicle’s driving behavior. Somebody has a name and an address.

Section 12 provides for an operator licensing scheme for no-user-in-charge operation, which is the regulatory handle on fleets running without anyone responsible inside the vehicle.

Section 47 is titled “User-in-charge not liable for manner of driving.” That heading is the heart of it. When the vehicle is lawfully driving itself, the person in the seat is not liable for how it drives. They may still have duties, around insurance or roadworthiness or taking over when properly asked, but the manner of driving is not theirs.

Be careful not to overread this. The Act is a framework, and a great deal depends on secondary legislation, authorisation criteria and how regulators actually run the scheme. A statute on the books is not the same as a settled body of case law. But the structural choice is clear and it is worth understanding: the UK moved responsibility for the driving task away from the occupant and onto a named corporate entity, and required that entity to be approved in advance.

The United States is still assembling the answer

There is no federal statute doing what the UK Act does. What exists instead is a layered patchwork.

State law covers testing and deployment. The National Conference of State Legislatures records 29 states plus Washington D.C. having enacted autonomous vehicle legislation, with governors in Arizona, Delaware, Hawaii, Idaho, Illinois, Maine, Massachusetts, Minnesota, Ohio, Washington and Wisconsin acting by executive order. Some of those statutes set minimum liability coverage, with figures as high as $2,000,000 appearing in the requirements. The categories states actually legislate are testing permits, insurance minimums, licensing exemptions and local preemption, not a reallocation of responsibility for the driving task.

Product liability does the rest of the work. A plaintiff alleging that software drove badly is generally arguing design defect, manufacturing defect or failure to warn, which are the same doctrines that apply to a brake caliper. That framing has real consequences, because a software design-defect case requires showing a safer alternative design was feasible, which is a very different exercise from showing a driver ran a red light.

Federal crash reporting and recalls supply the regulatory pressure. NHTSA’s crash-reporting obligations produce the data that made the Insurance Institute’s July 2026 analysis possible in the first place, and recalls remain the fastest formal lever. Rivian’s October 2, 2026 R2 recall over poorly tightened battery packs is an ordinary mechanical example, and the same mechanism now applies to software fixes delivered over the air, which is how several automated-driving defects have been addressed.

And then there is the timeline problem. The Lyft driver-classification case is not an autonomous vehicle case, but it is an instructive one about pace: the California Labor Commissioner filed in August 2020 and the $272.5 million settlement was announced on October 1, 2026. Six years. Liability questions involving large transportation platforms resolve on a timescale your kid will reach adulthood inside.

Who is likely responsible, scenario by scenario

Scenario	Most likely responsible	What is unsettled
Level 2 car, driver supervising, rear-end crash	The human driver	Whether a poorly designed attention warning shifts any share
Level 2 car, driver was hands-off, fatal crash	The human driver remains responsible for the driving task	Whether the system’s design contributed; this is what investigations examine
Driverless robotaxi, no occupant control	The operator and, through product liability, the manufacturer	How fault is apportioned; insurer subrogation; state-by-state variation
UK self-driving vehicle, lawfully engaged	The authorised self-driving entity	How authorisation criteria will be applied in practice
Software update introduced a defect	Manufacturer, via recall and product liability	Whether an over-the-air fix satisfies the remedy obligation
Sidewalk delivery robot injures a pedestrian	Operator and its insurer	Largely open; sidewalk rules are state and municipal
Crash during a federally reported investigation	Undetermined until the investigation concludes	Everything; an open investigation assigns no fault

The last row is the one most often misread in public conversation. When a regulator opens an investigation, the regulator is requesting information. That is all. It is not a determination that the system failed, that a defect exists, or that the manufacturer did anything wrong. Treating an investigation as a verdict is the single most common error in reporting on this subject, and it is worth correcting at your own dinner table before it shows up in your kid’s homework.

What a family should actually do about it

Ask your insurer the Level 2 question in writing

Call and ask how your policy treats a crash that occurs while a driver-assistance system is engaged, and get the answer by email. Most policies treat it as ordinary driving, which is consistent with the law, and knowing that in advance prevents a bad surprise. If you have a teen driver, pair this with our walkthrough of car insurance basics to teach before they drive.

Know the difference between a rideshare trip and a robotaxi trip

In a rideshare vehicle there is a human driver whose conduct is at issue, plus a platform whose classification of that driver has been litigated for years. In a driverless vehicle there is no driver and the operator stands in that place. Those are different legal structures behind an identical-looking app, and a teenager old enough to book either should understand which one they are in. We lay out the decision in robotaxi teen riders: what parents decide first.

Document, do not diagnose

If you are ever in an incident involving an automated system, record what the system was doing, what alerts sounded, what the screen displayed and the exact time. Photograph the instrument cluster if it is safe to do so. Resist the urge to state a conclusion about what the software did, in writing or to anyone. Facts are useful later; theories are not.

Teach the three-word distinction

Grounding, investigation, recall. Three different events, three different meanings. A voluntary grounding is a company choosing caution, as when Zoox pulled its Atlanta test fleet in September 2026 over a workplace exposure concern. An investigation is a regulator asking questions. A recall is a defect determination with a required remedy. We sort them out in how autonomous vehicle recalls actually work.

What not to do: don’t assume the brochure allocates liability

No amount of marketing language transfers legal responsibility from a supervising driver to a manufacturer. Only a statute or a court does that, and in the United States neither has yet done it for consumer vehicles. If a salesperson implies otherwise, that is a reason to ask for the automation level in writing, not a reason to relax.

What to Watch For Over the Next 3 Months

• Week 4: Watch whether the NHTSA investigation into Comma’s hands-off system escalates from a preliminary evaluation to an engineering analysis. That is a procedural step with real meaning, and it is still not a finding of fault.
• Month 2 red flags: Any state legislature introducing a bill that reassigns the driving task away from the human for consumer vehicles. That would be the first American move toward the UK’s structure, and it would change what your insurance conversation should sound like.
• Month 3 self-check: Ask whether every driver in your household can name the automation level of the car they drive most. If anyone answers with a trade name instead of a level, they do not currently know their own legal position.

Frequently Asked Questions

If my car’s lane centering causes a crash, is the manufacturer liable?

Not automatically. In a Level 2 system the supervising driver is responsible for the driving task, and a claim against the manufacturer would generally proceed as a product liability case alleging a design defect, a manufacturing defect or a failure to warn. That is a harder and slower argument than ordinary negligence.

Does the UK law mean nobody in the car is responsible for anything?

No. Section 47 of the Automated Vehicles Act 2024 removes liability for the manner of driving while the vehicle is lawfully driving itself. Other duties can still attach to the user-in-charge, and the authorised self-driving entity takes on the driving-task responsibility. The Act reallocates responsibility rather than deleting it.

Who pays if a driverless taxi hits my car?

In practice the operator’s insurer, with the operator standing in the place a driver would occupy, and with the possibility of a product liability claim against the manufacturer behind that. The apportionment varies by state, which is exactly the gap a federal framework would close.

Is an open NHTSA investigation a sign the company did something wrong?

No. It means a regulator is gathering information about a pattern it wants to understand. Many investigations close without a defect determination. Reading an investigation as a verdict is the most common mistake in coverage of this field.

Should I let my teenager use hands-off driving features?

The honest answer is that there is no measured safety benefit from the partial-automation bundle in IIHS’s comparison, and the legal responsibility stays with your teenager regardless. Those two facts together make a conservative setting the easy call during the learning years.

Will there ever be one clear rule?

Probably, eventually, and the UK has shown one version of what it looks like. The realistic near-term picture in the United States is continued state variation plus product liability, with recalls and federal crash reporting carrying most of the regulatory weight.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Parliament of the United Kingdom. (2024). “Automated Vehicles Act 2024” (2024 c. 10), sections 6, 12 and 47. legislation.gov.uk. https://www.legislation.gov.uk/ukpga/2024/10/contents
2. National Conference of State Legislatures. “Autonomous Vehicles.” https://www.ncsl.org/transportation/autonomous-vehicles
3. National Highway Traffic Safety Administration. “Automated Vehicles for Safety.” U.S. Department of Transportation. https://www.nhtsa.gov/vehicle-safety/automated-vehicles-safety
4. Insurance Institute for Highway Safety. “Advanced driver assistance.” https://www.iihs.org/topics/advanced-driver-assistance
5. California Department of Industrial Relations. (2026, October 1). “California Labor Commissioner’s Office announces record-setting $272.5 million settlement with Lyft.” https://www.dir.ca.gov/DIRNews/2026/2026-78.html
6. O’Kane, S. (2026, October 2). “Rivian issues R2 recall for poorly tightened battery packs.” TechCrunch. https://techcrunch.com/2026/10/02/rivian-issues-r2-recall-for-poorly-tightened-battery-packs/
7. O’Kane, S. (2026, September 23). “Comma’s hands-off driving tech under investigation after 2 fatal crashes.” TechCrunch. https://techcrunch.com/2026/09/23/commas-hands-off-driving-tech-under-investigation-after-2-fatal-crashes/
8. Wikipedia contributors. “Self-driving car.” https://en.wikipedia.org/wiki/Self-driving_car