Ask anyone about self-driving cars and the first question is the same: when there is no driver to blame, who pays? Motor insurance was built around the at-fault driver, so removing that driver collapses a century of assumptions about coverage. What now gates robotaxi expansion is authorisation and liability: what makes the vehicle legal, who can be held to account, and which regulator clears it to scale.
Take Nevada: it approved roughly 7,100 robotaxis for one county in one meeting, while the entire active US fleet is about 4,000 vehicles. Permits now outnumber the cars on the road. Regulation is a stack of gates: insurability, vehicle legality, then deployment permits, each a speed-versus-safety trade-off.
How do you assess the liability and insurance exposure of putting automated vehicles into service?
Remove the at-fault driver and the “who pays” question shifts from the motorist to the operator, manufacturer and software provider. To assess exposure you have to evidence data access, incident reporting and crash-outcome disclosure so insurers can underwrite the fleet, and settle liability allocation before any vehicle carries passengers.
With no single driver to point at, claims handling and subrogation have to be split across the parties who build, own and operate the vehicle, and what data and disclosure evidence makes a fleet insurable moves to the front. That is the first of the regulatory gates you clear.
Insurability has to be evidenced. Underwriters price against your disclosure record, and the industry still lacks enough data for sound risk modelling. Camera-only stacks act as a black box insurers struggle to analyse, while LiDAR-plus stacks and the vision-language-model approach Wayve pioneered aim for explainability insurers can price.
Coverage follows. Consolidated programmes wrap owner, operator, manufacturer and software provider in a single policy, the model Uber launched and where the market is heading. Vertically integrated players like Tesla, Amazon and Waymo lean toward self-insuring instead, which is the aggregator versus vertical-integration bet in miniature. Waymo’s own claims analysis reports 88% fewer property damage and 92% fewer bodily injury claims than the overall human population, though those are the company’s own figures. Insurability is a deployment precondition, sitting inside the per-mile economics before a ride is sold.
Why does a controls-free vehicle need an NHTSA exemption before it can carry paying passengers?
A vehicle with no steering wheel or pedals cannot comply with Federal Motor Vehicle Safety Standards written for human drivers, so it needs an NHTSA Part 555 temporary exemption before carrying paying passengers. NHTSA grants one only where it finds equivalent overall safety, and it caps the fleet while it gathers real-world data.
Even before a fleet can be insured, the vehicle itself has to be legal. In plain terms, seven of the eight standards Zoox sought relief from assume a human driver and hand-operated controls, covering wipers, turn signals, mirrors, sun visors, a brake pedal and air-bag warning labels. None of that has a job once an automated driving system takes over, but the law still demands it, so a controls-free car cannot comply with the rulebook as written. Zoox won the first passenger-carrying grant, Exemption No. 2026-01, effective July 2026 through July 2028, after an August 2025 application and 119 public comments. The grant is capped at 2,500 vehicles per 12 months and arrives with operational authorisations: adjustable conditions covering the operational design domain, remote assistance, workforce location and latency, and data access. Breach them and NHTSA can revoke the exemption. It is a temporary bridge, so reaching real scale still depends on rulemaking and updated standards.
Tesla’s Cybercab is the live test of the other path. Tesla self-certified the controls-free Cybercab as compliant, and NHTSA has opened an audit query and a Special Order asking how exactly it certified the car. DOT‘s AV 5.0 strategy commits to faster, more predictable exemption processing and to modernising the standards. Whether you seek an exemption or self-certify, you are still operating inside the regulatory terrain that decides how far your fleet can go.
US self-certification versus Europe’s pre-market approval: which gets robotaxis to scale faster?
The US lets manufacturers self-certify FMVSS compliance and deploy quickly, concentrating risk in post-market oversight. Europe requires pre-market type approval plus national and sub-national permits, which is slower but more uniform. Neither wins outright: speed trades against uniformity, and “safer” turns on process, disclosure and crash outcomes.
The vehicle question scales into a regime question. Under self-certification, a manufacturer attests its vehicle complies with the standards and deploys without prior government approval, with NHTSA’s recall, defect-investigation and crash-reporting authority as the backstop. There is no single national permit. The core difference from Europe is timing: the US checks after deployment, Europe checks before.
Europe’s approval is assembled piece by piece. A patchwork of national testing permits and approval procedures has made cross-border launches difficult, which is why 18 transport ministers signed a declaration for coordinated testing. Spain’s first national Level 4 permits to WeRide, Uber and AVOMO anchor a 20-vehicle Madrid fleet, while London’s staged Uber and Wayve launch starts with safety drivers on board.
Inside the US, two poles sit side by side. California stacks two gates, a DMV deployment permit plus a CPUC authorisation, which enabled Waymo’s 18-county expansion and up to 1,000 vehicles and shows the aggregator model’s reach across permitting regimes. Arizona is more permissive, folding robotaxis into its existing transport network rules. Which pole is safer is a separate question from which is faster.
Australia sits between the two regimes. AV rules are state-led, the Automated Vehicle Safety Law is now expected in 2027, and over 80% of Australians surveyed express uncertainty about liability. It is a useful reference point in the broader scale-up story, and once a regime clears your fleet, what real scale demands operationally is the next problem.
Reading any robotaxi announcement
The “who pays” question you walked in with is the same question that decides whether any robotaxi gets on the road. The three gates resolve into one structure: insurability, vehicle legality, then deployment permits, and liability is the thread running through all of them. No underwritable risk model means no fleet, however good the software.
Judge every deployment announcement by its gates. Who holds liability? Is the vehicle legal? Which regime cleared it to scale? The answers are still being written in exemptions, permits and policies, and they will decide who scales and how fast.
Frequently Asked Questions
What does “controls-free” mean for a robotaxi?
A controls-free robotaxi has no steering wheel, pedals or other manual controls, so a human cannot take over. The automated driving system (ADS) performs the entire driving task within its operational design domain. Because Federal Motor Vehicle Safety Standards assume a human driver and hand-operated controls, a controls-free design falls outside them and needs an NHTSA exemption before it can carry paying passengers.
How long does an NHTSA exemption last, and what happens when it expires?
A temporary exemption under 49 CFR Part 555 runs for a fixed period, such as Zoox’s Exemption 2026-01, effective July 2026 through July 2028. When it lapses, the operator must have a renewal or a permanent pathway in place. Exemptions are a bridge, so lasting change depends on FMVSS modernisation and rulemaking rather than one-off grants.
What happens when a robotaxi fleet reaches the 2,500-vehicle exemption cap?
The cap of 2,500 vehicles per 12 months limits how far an exemption can carry a fleet. Once an operator approaches it, further growth depends on a broader authorisation, such as updated FMVSS or a new regulatory pathway. The cap is deliberate: it keeps early controls-free deployment contained while regulators study real-world safety data.
Is an operational authorisation the same as an exemption?
No. An exemption makes a controls-free vehicle legal by granting relief from specific standards, while an operational authorisation is the set of adaptable conditions NHTSA attaches to that grant. Those conditions cover the operational design domain, remote assistance, workforce location and latency, and data access. Together they turn a static exemption into living permission that regulators can adjust over time.
What is an operational design domain (ODD)?
An operational design domain (ODD) is the specific set of conditions in which an automated driving system is designed to work safely, such as defined streets, weather, speed limits and times of day. An operator cannot simply claim its system works everywhere; it must declare its ODD and prove the vehicle performs within it. Regulators tie permits and exemptions to that boundary.
What’s the difference between a permit, a licence and an authorisation?
In practice the terms are used interchangeably for the state or national gate that lets a robotaxi service operate, so readers should not assume a strict legal distinction. The important split is between federal vehicle approval (self-certification in the US) and the deployment approval layered on top. When comparing markets, focus on which body grants permission and what conditions it attaches.
Do US robotaxi operators need permission from every state they cross?
Yes. The US has no single national deployment permit, so operators must clear each state’s rules, and often city or county ones as well. California, for example, requires approval from both the CPUC and the California DMV, while Arizona takes a lighter-touch approach. That patchwork is why scaling a US robotaxi service is a state-by-state exercise.
Why do insurers care whether a self-driving car uses cameras or LiDAR?
Because insurability depends on evidence an underwriter can interrogate. A camera-only stack can behave like a black box, making it hard for insurers and regulators to reconstruct why a crash happened. LiDAR-plus and vision-language-model systems, such as Wayve’s, aim for explainability, which makes crash causation clearer. Easier-to-explain systems are easier to price and insure.
Is Europe’s slower approval process actually safer?
Not necessarily. Europe’s pre-market type approval is slower but more uniform, while the US self-certification model deploys faster and pushes risk into post-market oversight through recalls and defect investigations. Neither process guarantees safer outcomes on its own. Safety turns on process, disclosure and real crash results, so speed and safety are separate questions.
What happens if no insurer will cover a robotaxi fleet?
The fleet cannot lawfully operate. Insurability is a deployment precondition rather than a post-crash detail, so an operator that cannot evidence data access, incident reporting and crash-outcome disclosure will not secure cover, no matter how capable its software. That is why liability allocation across the operator, manufacturer and software provider is settled before vehicles carry passengers.
Are there any robotaxis operating in Australia yet?
Australia is still in the trial and rule-setting stage rather than full commercial service. AV rules are state-led and sit somewhere between the US and European models, without a single national framework. That makes Australia’s position a useful comparison for readers watching how the two dominant regimes scale, but it is not yet a commercial robotaxi market.
What is SAE J3016 Level 4, and why does it matter for permits?
SAE J3016 is the standard that defines driving automation levels, and Level 4 means the system can drive itself without human intervention within a defined operational design domain. Most commercial robotaxis aim for Level 4 rather than full Level 5. Regulators use that level to set expectations for permits, because it fixes where and when the vehicle may operate.
