When software steers, who pays for the crash? The UK has legislated an answer, Mercedes has volunteered one, and the messy middle of driver assistance still belongs to the driver.
Motor insurance has rested on one assumption for a century: a human was driving, so a human is liable. Automation dissolves that assumption gradually — lane-keeping here, automatic braking there, hands-off highway modes in a few markets — and the legal system is answering the resulting question in pieces. The pieces are worth reading, because they sketch where motor liability is heading everywhere, including in markets that have not legislated yet.
The direct answer, as far as one exists today: for genuinely automated driving, liability is migrating from the driver to the manufacturer, with the insurer paying first and recovering afterwards. For driver assistance — which is what almost everyone actually owns — the driver remains liable, and that is precisely where the disputes live.
What the UK wrote into law
The clearest legislative answer so far is British. The Automated and Electric Vehicles Act 2018 made the insurer directly liable to victims of an accident caused by an automated vehicle driving itself — the injured party claims against the motor policy exactly as they always did, and the insurer then pursues recovery from whoever was actually at fault, typically the manufacturer.
The Automated Vehicles Act 2024, which received royal assent in May 2024, completed the structure. When an authorised self-driving feature is engaged, the person in the seat — the user in charge — is not responsible for how the vehicle drives, in either criminal or civil law. Responsibility shifts to an authorised self-driving entity, normally the manufacturer or developer, which must meet standards of good repute and financial soundness so there is always a well-resourced defendant to recover from.
The design is deliberate: victims are paid quickly through the familiar insurance channel, and the argument about whether software or human caused the crash happens afterwards, between institutions that can afford to have it.
Pay the victim first through insurance; let insurers and manufacturers argue about the software later.
What manufacturers have volunteered
Legislation is one path; contractual promise is another. Mercedes-Benz stated in 2022 that it accepts legal responsibility for accidents that occur while its Level 3 Drive Pilot system is engaged and operating within its design conditions. That single announcement did a lot of quiet work: it conceded, from inside the industry, that when the system is driving, the maker of the system owns the outcome.
The caveats matter as much as the promise. The commitment applies within the system's operating envelope — specific roads, speeds and conditions — and disputes will turn on whether the system was truly engaged and operating as designed at the moment of impact. Which is why the most valuable component in an automated vehicle, from a liability standpoint, is the data recorder.
The messy middle: assistance is not automation
Nearly everything on the road today is Level 2 or below: the system assists, the human supervises, and the human is liable. Courts in several jurisdictions are working through crashes involving assistance systems marketed in language that suggested more capability than the law recognises, and the recurring question is not whether the driver was in charge — legally they were — but whether the product's design and marketing invited them to believe otherwise.
For insurers this middle zone cuts in two directions at once. Assistance systems demonstrably reduce crash frequency, which is good for claims. But they also concentrate ambiguity: every incident involving lane-keeping or adaptive cruise now carries a potential dispute about system behaviour, sensor state and driver attention, and resolving it depends on telemetry the insurer does not control.
What this means in the Gulf, and for distribution
Gulf markets, Saudi Arabia included, have not yet legislated automated-driving liability, but the vehicles arriving in showrooms carry the same sensors and software as everywhere else, and regional regulators tend to draw on frameworks proven elsewhere. The UK's structure — insurer pays first, recovers from the manufacturer, driver protected when the system is genuinely in charge — is the template most likely to travel, because it keeps victim compensation simple while the technical argument proceeds offstage.
For anyone distributing motor cover digitally, two practical consequences follow. First, vehicle data becomes pricing data: what assistance hardware a car carries will increasingly matter to its premium, which rewards distribution channels that capture accurate vehicle specifications at the point of quote. Second, product wording will evolve — expect policies to define system-engaged scenarios explicitly rather than relying on a century of human-driver assumptions.
The honest limit of this piece: the case law is young, most of it is unsettled, and no market has yet run a mature automated-vehicle claims regime at scale. What exists today is a legislative sketch and a manufacturer's promise. Both point the same way — liability follows control — but the detailed map is still being drawn, filing by filing.