Robotaxis and liability: who pays when a driverless car crashes

When a driverless car crashes, the company that operates it, not a person in the front seat, is normally the party whose insurance and legal responsibility are in play. Several states now say so in statute: when the automated driving system is engaged, the system is treated as the operator, and the service must carry insurance as a condition of running. What differs from a normal crash is who the claim is against, what the passenger agreed to in the app, and how much of the evidence sits on the company's servers. This article describes how that works as of November 2026, using California, Florida and Texas as examples. Other states differ.

Who counts as the driver when no one is driving

Ordinary crash law starts with a driver whose carelessness can be measured against a standard. A driverless vehicle has no such person, so states have had to name someone.

Florida's statute says that when the automated driving system is engaged, the system "shall be deemed to be the operator" of the vehicle, whether or not a person is physically present (Fla. Stat. 316.85, 2026). Texas's 2025 law, SB 2807, uses similar language: when an automated driving system is engaged, it is the operator, and traffic citations go to the vehicle's owner or, for a vehicle running under a state authorization, the authorization holder (SB 2807, enrolled text, 89th Legislature). The same bill says its compliance requirements begin 90 days after the state agencies adopt implementing rules, so check where Texas stands before assuming any rule is in force.

California takes a permit route. A manufacturer needs a Department of Motor Vehicles permit to test or deploy driverless vehicles, and a company that charges the public for rides needs a separate permit from the California Public Utilities Commission (CPUC). Since July 1, 2026, a peace officer who sees an alleged traffic violation can issue a "Notice of Autonomous Vehicle Noncompliance" to the manufacturer, and the manufacturer must forward it to the DMV within 72 hours, or 24 hours if the officer marks it for priority review (DMV memo AVIM 2026-002).

None of this answers the civil question by itself. Naming the system as the "operator" for traffic law tells you who gets the ticket. Who owes damages after an injury is decided by negligence and product-liability rules, covered below.

What insurance robotaxi services must carry

The insurance floor for a robotaxi service is set by state law, and it is not the same as the minimum limits on a private policy. Here are three examples.

Two points follow. First, the large policy limits belong to the company, so a passenger or another motorist hurt by a robotaxi is usually dealing with a corporate insurer or a self-insured operator, not an individual's personal auto policy. Second, a minimum is a floor. A catastrophic injury can exceed a $1 million or $5 million limit, as it can with any policy, and then the claim turns on the company's assets and its other coverage. For how this compares with human drivers on Uber or Lyft, see our guide to Rideshare insurance for Uber and Lyft drivers and the article on state laws on rideshare insurance requirements.

Negligence versus product liability

A person injured by a driverless car can usually frame the claim two ways, and the framing decides what has to be proved.

Negligence asks whether the operator failed to use reasonable care. When a human driver runs a red light, the question is simple. When software does it, the inquiry moves to how the system was designed, tested, maintained, monitored by remote staff and permitted to operate in that place and weather. The operating company is the natural defendant: it chose the route area, the software version and the supervision.

Product liability asks whether the product was defective. Cornell's Legal Information Institute summarizes the usual categories as manufacturing defect, design defect, marketing defect (inadequate warning) and breach of warranty, and notes that strict liability does not require proof of intent or knowledge: the plaintiff shows a defect existed when the product left the defendant's hands and that it caused the injury. For a robotaxi, a design-defect theory might say the software was unreasonably unsafe when a safer alternative existed. Whether a given software behavior counts as a defect is a question for experts and courts.

In practice a plaintiff's lawyer may plead both against the operating company, and against the vehicle maker and component suppliers if they are different businesses. Which theory wins is for a court or a settlement to decide; a licensed attorney in your state can say which applies to a specific injury, and deadlines for suing differ by state (see statutes of limitations for car accident claims).

If you are a passenger: terms of service and your other coverage

Riding in a robotaxi means accepting terms in an app, and those terms matter after a crash. Waymo's published terms (last updated September 29, 2026) are a useful example of what to look for (Waymo Terms of Service).

• Arbitration and class waiver. The terms require most disputes to go to individual arbitration before JAMS and waive jury trials and class actions. There is a 45-day informal-resolution period before arbitration can be filed, and a rider can opt out of arbitration within 30 days of first accepting the terms by emailing the address the terms list. A few claims, such as intellectual property claims and individual claims over sexual misconduct by a Waymo employee or agent, are carved out.
• Limits on damages. The terms cap Waymo's liability for most claims at the amount you paid for the ride and exclude consequential damages, which would make little sense for an injury, so the same section lists exceptions. It says the limits do not apply to gross negligence, fraud or reckless or intentional misconduct, and not to personal injury or property damage to the extent caused by Waymo's negligence or a defect in its services, unless the rider's or a third party's conduct caused the harm.

Reading the exceptions together, a rider hurt by a fault in the service is not left with a refund-sized remedy under these particular terms, but the dispute may still run through arbitration unless the rider opted out. Other operators write their own terms, so read the one you accepted.

Your own coverage can respond too. In a first-party claim, your health insurance, medical payments coverage or personal injury protection (required in some states and offered in others) may pay medical bills without waiting for anyone's fault to be decided. Then the insurer that paid may seek reimbursement from the responsible party, a process called subrogation. Our article on passengers injured in a rideshare crash: how claims work walks through the order of payments for human-driven rides, and much of it applies here.

If a human driver hits a robotaxi, or a robotaxi hits you

For a person in an ordinary car, the basic rules do not change. Fault is decided on the facts: who had the right of way, who followed too closely, who changed lanes. Your insurer handles a claim against you as usual, and a claim you make against the robotaxi's operator is a third-party claim, the same category as a claim against any at-fault driver. Our guide on how fault is determined in a car accident covers how insurers decide.

What is different is the evidence and the other side. The operating company's insurer or claims group will have vehicle logs and camera footage you do not. If the robotaxi was in the wrong, your insurer pays for your repairs under your collision coverage if you have it and then pursues the operator through subrogation, or you bring the claim directly. If you were in the wrong, the operator's insurer will pursue you the same way it would pursue any driver.

Evidence worth gathering before you leave the scene:

1. Photos of all vehicles, the lane markings, signals and the robotaxi's identifying marks and plate or fleet number.
2. Names and contact details of witnesses, and whether the robotaxi had passengers.
3. Your own dash camera footage, saved immediately before it loops over. See dashcam footage and scene evidence: what to capture and keep.
4. The company's reporting channel. Note any claims contact or support number the operator provides in the app or in the vehicle.
5. A written preservation request sent to the operator and its insurer soon after the crash, asking that sensor data, video and remote-assistance logs for the time and place be kept. A lawyer can word this properly.

The after-accident checklist and information form covers the general scene steps that apply to every crash.

Data recorders and how long the evidence survives

A robotaxi crash is documented unusually well, but the records belong to the company. California's deployment regulations require an "autonomous technology data recorder" that captures and stores sensor data for all functions controlled by the automated driving system for at least 30 seconds before a collision, in a read-only format, and requires the data to be preserved for three years after the collision (Article 3.8, as adopted). That is a manufacturer's duty to a regulator, not a right of the other driver to the file; getting it typically takes a request through insurers or a lawyer, and in litigation a subpoena.

Federal reporting adds another layer. Under NHTSA's Standing General Order, companies named in the order report certain crashes in which an automated driving system was engaged within 30 seconds of the crash. As described in a May 2026 notice in the Federal Register, an initial report is due within five calendar days for the more serious categories, such as a hospital transport, a fatality, an airbag deployment or a strike of a vulnerable road user, and for driverless systems also a tow-away; other incident reports follow monthly. NHTSA says it publishes incident report data monthly with redactions (Federal Register notice 2026-10363). Those reports let regulators and researchers see patterns, but they are summaries, not a substitute for the data in your own claim. Our article on event data recorders: what your car records and who can access it explains the conventional recorders in human-driven cars.

Two dated examples of how the mechanism plays out

A safety driver and a company, 2018 to 2019. In March 2018 an Uber test vehicle with a backup safety driver struck and killed a pedestrian in Tempe, Arizona. The National Transportation Safety Board voted in November 2019 that the probable cause was the safety driver's failure to monitor the road because she was distracted by her phone. It also criticized Uber's safety culture, and prosecutors had said earlier that Uber was not criminally liable (Insurance Journal, November 20, 2019). The lesson for liability is that responsibility can be spread: a human supervisor, a company's design choices and the pedestrian's own conduct were all examined. That case involved a test vehicle with a person who was supposed to intervene, which is different from a driverless commercial service today.

A reporting failure, 2023 to 2024. In October 2023 a Cruise robotaxi in San Francisco dragged a pedestrian about 20 feet before coming to a stop, seriously injuring the pedestrian. California suspended Cruise's license to operate, Cruise recalled its driving software and NHTSA opened a defect investigation. At the end of September 2024 NHTSA announced a consent order with a $1.5 million civil penalty because two reports about the incident omitted post-crash details required by the agency (Insurance Journal, October 1, 2024). The relevance here is evidentiary: the fight was over what the company's records showed and when regulators saw them.

Safety claims, and why they do not settle liability

Supporters point to claims data. A December 2024 study by reinsurer Swiss Re, published with Waymo, compared liability claims over 25.3 million driverless miles against human-driver baselines and reported an 88% reduction in property-damage claims and 92% in bodily-injury claims; Waymo had nine property-damage claims and two bodily-injury claims, and the two injury claims were still open (Waymo blog, December 19, 2024). Treat that as one company's data, analyzed with an insurer and released by the company, using claims as a proxy for at-fault crashes. It says nothing about whether any single crash was the operator's fault, and it covers only one operator in a few cities.

Privately owned self-driving cars: open questions

Robotaxi law is simpler than the law for cars people own, because a single company controls the vehicle, the software and the insurance. For privately owned vehicles, several questions are unsettled:

• When a system that is not fully driverless is steering, does the owner, the maker or both answer for a crash? Our article on Driver-assist crashes: who pays when Autopilot-style features were on covers the current situation for driver-assist features, where the human is still responsible.
• Will owners buy ordinary personal auto policies that treat the manufacturer as the main target of claims, or will makers carry insurance themselves? Texas's SB 2807 requires coverage "equal to or greater than" the state's required amount for the vehicle's use, which is a floor and says little about who ultimately pays.
• How will insurers price a policy for a car whose crash risk depends on a vendor's software updates? See connected-car data, data brokers and your premium for the data side.

For now, if you own a car with advanced driver-assist features, read your policy and ask your insurer how it treats a crash while a feature was engaged. Your obligations as the driver do not disappear because the feature is called "self-driving" in marketing.

Common questions

Can I sue a robotaxi company if I am hurt as a passenger?

Possibly, but read the terms you accepted. Many operators require arbitration and give a short window to opt out, as the Waymo terms above do. The same terms may preserve full liability for injuries caused by the company's negligence or a defect. A licensed attorney in your state can tell you how the arbitration clause would be treated there.

Whose insurance pays if a robotaxi hits my parked car?

The operator's insurer is the starting point, since the operator is responsible for the vehicle. You can also claim on your own collision coverage and let your insurer recover from the operator, though you would owe your deductible first. The first-party vs third-party claims guide explains the trade-off.

Does a robotaxi crash raise my premium?

It can if your insurer finds you at fault, as with any crash. If the operator's records show the robotaxi was at fault, that finding helps you show you were not. Ask for the written fault decision.

Is the safety driver or the company responsible in a test vehicle?

It depends on the facts and the state. In the Tempe case discussed above, regulators found fault with both the driver and the company's practices. For a specific test-vehicle crash, ask a lawyer.

Key takeaways

• States such as Florida and Texas treat an engaged automated driving system as the "operator," and California regulates through DMV and CPUC permits; the rules vary, so check your state's insurance department or the state car insurance requirements lookup tool.
• Robotaxi services carry corporate coverage set by state law, such as $1 million primary liability in Florida and $5 million in California's programs; those are floors, not caps on liability.
• A claim can be framed as negligence by the operator or as product liability for a defect, and often both are pleaded.
• Passengers should read the arbitration, opt-out and damage-cap terms they accept, and should note that some terms keep liability intact for company negligence or defects.
• After a crash involving a robotaxi, photograph the scene, save your own video, and ask in writing that the operator preserve its data; a recorder may keep it for years, but it is not yours to demand without a claim or a lawyer.