For decades, researchers have agreed on one thing: humans are the weak link. According to the U.S. National Highway Traffic Safety Administration, 90 percent of crashes are caused by human error. That includes everything from glancing at a phone to misjudging a turn or driving under the influence.
Self-driving cars don’t drink. They don’t get tired. They don’t take risks to shave a minute off the commute. This is the promise of autonomy: reduce human involvement, and you reduce the lion’s share of crashes.
But there’s a caveat. Humans also bring creativity and adaptability. We make eye contact with pedestrians at crosswalks. We anticipate the hesitation of a nervous driver in the next lane. These subtle interactions are areas where AVs, despite their sensors and algorithms, still struggle.
The question is whether those limitations outweigh the benefits of removing human distraction, fatigue, and poor judgment. Early studies suggest they don’t.
The Safety Problem We’re Trying to Solve
Every year, more than 40,000 people die on U.S. roads and thousands more in the UK. The vast majority of these crashes are caused by human error, mistakes like speeding, drink-driving, or glancing at a phone. Regulators estimate between 88 and 94 percent of collisions come down to human decisions.
This is why autonomous driving matters. Even a modest improvement in safety could save hundreds of thousands of lives over a generation. RAND modelling suggested that if self-driving cars were just ten percent safer than humans, they could prevent 600,000 U.S. fatalities over 35 years.
Where AVs Shine and Where They Struggle
One of the most comprehensive studies to date, published in Nature Communications in 2024, compared more than 2,100 AV-related crashes with 35,000 human-driven accidents. The results paint a nuanced picture.
AVs were far less likely to cause certain kinds of collisions. Rear-end crashes, among the most common on our roads, dropped significantly when cars were under automated control. Broadside collisions too were far less frequent. The same held true in poor weather: radar and lidar sensors allowed AVs to outperform humans in rain and fog, conditions where drivers often falter.
Yet the study also revealed where machines fall short. At dawn and dusk, when light conditions are constantly shifting, AVs were more than five times as likely to crash compared to human drivers. Turning was another problem: self-driving cars were nearly twice as likely to have accidents in junctions or during turning manoeuvres.
The takeaway is clear. AVs excel at the routine and predictable, such as steady motorway driving or lane-keeping, but remain vulnerable in complex, human-heavy scenarios.
On the Ground: Waymo’s 25 Million Miles Study
Academic findings only go so far; what matters most is real-world evidence. Waymo, Google’s self-driving spin-off, has been quietly amassing one of the largest autonomous driving datasets in history, and the numbers are now public. Working with Swiss Re, one of the world’s leading reinsurers, Waymo analysed liability claims across 25.3 million miles of fully driverless operation in Phoenix, Los Angeles, San Francisco, and Austin.
The results were hard to ignore. Compared to the average human driver, Waymo’s system cut property damage claims by 88% and bodily injury claims by 92%. Even when stacked against newer cars loaded with advanced safety tech ,automatic braking, lane assist, blind spot monitoring, the reductions were still close to 90%. In practical terms, across all those millions of miles, only two bodily injury claims were filed, both still under review.
But the Swiss Re data is only part of the picture. In May 2025, Waymo released a complementary study, peer-reviewed and accepted by the journal Traffic Injury Prevention, that zoomed in on specific crash types. The results showed the Waymo Driver wasn’t just avoiding fender-benders — it was reducing the most dangerous accidents on the road:
- 92% fewer pedestrian injury crashes compared to human drivers.
- 82% fewer cyclist and motorcyclist injury crashes.
- 96% fewer injury-involving intersection crashes, where NHTSA data shows humans are most at risk.
- 85% fewer crashes with suspected serious or worse injuries, pointing to life-saving benefits, even though these events are fortunately rare.
Taken together, these results tell a compelling story: autonomous driving, at least in Waymo’s case, isn’t just safer overall — it’s especially better at protecting the most vulnerable people on the street.
Why Public Trust Still Lags
And yet, public opinion has not caught up with the data. Surveys in both the US and UK show widespread unease. In America, AAA found that 68 percent of people do not feel safe in self-driving cars, while trust in the technology dropped to just nine percent. In the UK, research by the Institution of Mechanical Engineers showed that seven in ten people would be uncomfortable riding in a car without human controls.
This is not purely rational. Trust is shaped by stories, and stories are shaped by headlines. The quiet millions of safe miles logged by Waymo rarely make news. What does make headlines are the exceptions: Uber’s fatal crash in 2018, Cruise’s pedestrian-dragging incident in 2023, or Tesla collisions involving drivers who misunderstood “Autopilot”.
The result is a perception gap. The reality is that AVs are already safer in many scenarios, but the public mostly remembers the rare tragedies. Until perception catches up with performance, adoption will lag.
The Policy and Insurance Puzzle
Technology alone does not make roads safer. It needs to be backed by clear rules and fair accountability.
The UK took a major step forward in 2024 with its Automated Vehicles Act, which formally defined liability and safety standards based on the SAE Levels of automation. For the first time, it was written into law who is responsible if a Level 3 or 4 system crashes: sometimes the driver, sometimes the manufacturer.
For insurers, this transition is particularly tricky. At Level 2, responsibility is clear: the driver is always accountable. But at Level 3, when the car can take full control in defined conditions, liability becomes blurred. If a crash occurs, who pays? The driver, who was not supposed to be driving at the time? The automaker? The software developer?
This grey zone is exactly why companies like Zego are developing av insurance products that evolve with automation. Zego already leads in providing flexible motor insurance for commercial drivers, and we see autonomous vehicle insurance as the next frontier. Our goal is to make sure accountability moves smoothly from human to machine, ensuring that drivers, businesses, and fleet operators are never left in legal limbo as the technology advances. By bridging the gap between regulation and real-world use, Zego is helping to build the trust and infrastructure that self-driving cars need to thrive.
Where We Really Are
For all the hype about robotaxis and driverless futures, most cars on the road today are still very much in the early stages of autonomy.
The majority of new vehicles are at Level 1 or 2, with systems such as adaptive cruise control or lane-keeping that require constant driver supervision. A handful of cars, notably some Mercedes models, now offer Level 3 capability in controlled motorway conditions.
Waymo and Cruise are pioneering Level 4 robotaxis, operating in geofenced city zones. These cars can drive themselves without human input, but only in mapped areas and favourable weather.
Level 5, true autonomy everywhere, all the time, with no steering wheel or pedals, remains a long-term goal. Some experts believe it could take decades, if it ever arrives at all.
So, Are Self-Driving Cars Safer?
The fairest conclusion is that yes, in many cases, they already are. Machines do not text while driving. They do not fall asleep at the wheel. They do not drink or get distracted. The weaknesses that remain, such as dawn light, sharp turns or unpredictable junctions, are engineering challenges, not fatal flaws.
And the trendline is unmistakable. With every additional million miles, the gap between human and machine safety performance grows. The machines are learning, improving, and getting safer at a pace humans cannot match.
It is worth remembering that humans did not become safe drivers overnight either. It took decades of regulation, licensing and infrastructure design to make our roads as safe as they are today. Self-driving cars are on their own learning curve, one that is already producing fewer crashes, fewer injuries, and safer streets.
The road ahead will still have bumps. But the destination, fewer collisions, more consistent safety, and insurance frameworks that keep pace with automation, is one worth striving for.