In the UK, the question of who is liable when a driverless car crashes is no longer theoretical. With the introduction of the Automated Vehicles Act 2024 (AV Act), the legal groundwork has been laid for a future where the human driver is no longer at fault, because they’re no longer in control. But even as the framework is set, real-world deployments are raising new questions about how risk is priced and responsibility assigned.
"This is probably the question I am most commonly asked," said Ben Gardner (pictured), partner at law firm Shoosmiths, referring to the long-debated issue of liability in autonomous vehicle (AV) accidents. "If you have a driverless vehicle that is in an accident, where does liability sit?"
According to Gardner, the AV Act creates two new legal entities to answer that question: the Automated Self-Driving Entity (ASDE), typically the car manufacturer, and the No-User-In-Charge operator (NUIC), which might be a mobility service provider like Waymo or Uber. "The Act essentially creates those two new legal actors to which liability will now attach," he said.
Rather than relying on courts to untangle post-crash blame, these entities will generally shoulder initial responsibility. The goal is swift compensation and clearer accountability as AVs become more common on UK roads.
While the new framework offers clarity on paper, its practical impact on insurance exposure remains nuanced.
"If there is no human liability, there should always be a third party against which the insurer can claim," Gardner said. This could streamline subrogation processes, offering more predictable recoveries for insurers. But the upside is tempered by unresolved variables.
Premiums might fall if AVs reduce accidents as promised. Yet Gardner cautions that the vehicles themselves are laden with costly tech. "If those vehicles are involved in an accident, the cost of repair and remediation may well be higher than for a conventional car."
The biggest challenge is data. Without real-world accident statistics, insurers are being asked to price risk ahead of clarity. "There is still a huge lack of data and learning at the moment," he said. "Insurers are having to take a bit of a punt."
That risk was underscored recently by a series of incidents involving robotaxis in North America. Initially praised for their caution, some vehicles – such as those operated by Waymo – have been reprogrammed to drive more assertively in traffic. That shift has led to illegal U-turns, close calls with pedestrians and cyclists, and even an incident involving a neighborhood pet. These behavioural changes challenge assumptions about AV safety and raise new underwriting concerns.
“In the short term there is no guarantee that AVs are a lower-risk proposition but data from trials and deployments demonstrate clear medium to long term safety benefits,” Gardner said. That said, the leap from test environments to real-world traffic introduces variables that current models aren’t fully prepared to quantify.
When collisions do happen, a digital trail will tell the story. Gardner expects every AV to be fitted with event data recorders capable of pinpointing software glitches, sensor failures, or telecoms issues. These black boxes will help determine fault and allow liable parties to reclaim damages through their supply chain.
The AV Act also establishes a new independent accident investigation branch with sweeping powers. "It can be a criminal offence for the ASDE or the NUIC to mislead or withhold information," Gardner said. Access to vehicle and sensor data will be critical, but contentious.
"ASDEs or NUICs are quite reticent because there is concern around access to intellectual property and trade secrets," he said. In response, the industry and government are expected to develop formal data-sharing protocols that balance transparency with commercial protection.
Despite the legislative strides, much of the AV Act remains skeletal. "There are scores of regulations that now need to be created to bring it to life," Gardner said, covering everything from cyber security to licensing.
With full implementation not expected until late 2027, insurers and operators face a long stretch of uncertainty. Grey areas persist, such as liability when a vehicle owner refuses to install critical software updates. "There are a few of those 'what if' questions that have not yet been answered," Gardner said.
Still, he believes the UK is striking a reasonable balance. "We are taking a fairly sensible approach. The ingredients are there," he said, pointing to growing interest from global AV firms eyeing the UK for advanced deployment trials.
For now, insurers are navigating a moving target. As recent incidents show, software behaviour updates can materially alter a vehicle’s driving profile, and insurers must be prepared to adjust accordingly. "They are having to collaborate with their customers and with industry to work out what 'suitable', 'appropriate' or 'good' looks like from an insurance perspective," Gardner said.
The era of human-free driving may be arriving, but the risk calculus is still under construction.
