Are driverless cars safe?

By Matthew Parish, Associate Editor

Wednesday 13 May 2026

The dream of the driverless motor car is as old as the motor car itself – the removal of human frailty from the act of driving, the promise of a machine that neither drinks, nor tires, nor grows distracted. In a world in which road traffic collisions kill more than a million people annually, overwhelmingly through human error, the appeal of automation is self-evident.

Yet the question of whether driverless cars are safe is not one that admits of a simple answer. The truth lies in a paradox: autonomous vehicles are, in certain measurable respects, safer than human drivers – and yet, when they fail, they can fail in ways that are both unexpected and, occasionally, catastrophic.

The statistical promise of automation

On paper the case for driverless vehicles is compelling. Systems such as those developed by Waymo and Tesla have demonstrated lower accident rates per mile than conventional driving in several datasets. One peer-reviewed study found an injury rate of roughly 0.6 incidents per million miles for autonomous systems, compared with 2.8 for human drivers – a striking reduction.

Similarly Tesla has claimed that vehicles using its Autopilot system experience far fewer crashes per mile than those driven without assistance.

More broadly, proponents argue that automation addresses the fundamental causes of most road accidents: distraction, intoxication, fatigue, and poor judgment. Autonomous systems do not text while driving, do not fall asleep at the wheel, and do not misjudge speed through impatience or bravado.

Indeed some studies suggest reductions of over 80 per cent in certain categories of serious crashes when autonomous systems are deployed under controlled conditions.

If one were to consider only these figures, the conclusion might appear obvious: driverless cars are not merely safe – they are safer.

The problem of context

However statistics conceal as much as they reveal. Autonomous vehicle safety data is often drawn from limited environments – well-mapped urban areas, favourable weather conditions, and fleets maintained to a high technical standard.

Critics note that such conditions do not fully replicate the chaotic variability of real-world driving. The fact that Waymo’s fleet has been involved in over a thousand reported incidents, albeit many minor and not necessarily its fault, illustrates that exposure increases complexity.

Moreover the growth in reported incidents – peaking in some datasets at over a hundred per month – reflects not only increased deployment but also unresolved challenges in mixed traffic environments where humans and machines must interact.

The central difficulty is not that autonomous vehicles cannot drive – it is that they must coexist with human beings, whose behaviour is often irrational, unpredictable and context-dependent in ways that are difficult to encode in software.

When systems fail

The most troubling aspect of driverless technology lies not in its average performance, but in its edge cases – the rare but consequential situations in which the system behaves incorrectly.

Recent reporting has highlighted instances in which autonomous vehicles have obstructed emergency responders, failed to interpret hand signals, or frozen in complex scenarios, thereby creating secondary hazards.

There have also been high-profile accidents involving autonomous or semi-autonomous systems, including fatalities. By late 2025, dozens of deaths had been associated with such technologies worldwide.

Even when not fatal, incidents can be unsettling. Vehicles have been reported to stall in intersections, misinterpret traffic signals, or collide with stationary objects – failures that, while statistically rare, undermine public confidence.

One particularly revealing pattern is the tendency of autonomous systems to struggle with ambiguity. Human drivers rely on tacit social cues – eye contact, gestures, informal negotiation at junctions. Machines, by contrast, require explicit inputs and predefined rules. When confronted with uncertainty, they may default to overly cautious behaviour, such as stopping abruptly, or to inappropriate action based on incomplete interpretation.

The psychology of trust

A further complication arises from human interaction with these systems. Semi-autonomous technologies, in particular, create a dangerous middle ground: the driver is neither fully in control nor fully disengaged.

This can lead to overconfidence. Drivers may assume that the system is more capable than it is, reacting too slowly when intervention becomes necessary. Analysts frequently identify this misplaced trust as a major contributor to accidents involving advanced driver assistance systems.

Fully autonomous vehicles, which remove the human driver entirely, avoid this specific problem – but introduce others, particularly concerning accountability. When a machine makes a mistake, responsibility becomes diffuse: is it the manufacturer, the software developer, or the operator of the fleet?

Catastrophe and perception

It is worth noting that catastrophic failures of driverless systems attract disproportionate attention. A single fatal accident involving an autonomous vehicle may receive global coverage, whereas the daily toll of human-driven accidents passes largely without remark.

This asymmetry distorts public perception. Humans tolerate a high level of risk from other humans, but expect near perfection from machines.

Yet this expectation is not wholly unreasonable. When technology is marketed as safer than human judgement, its failures are judged against that promise. A system that is statistically safer overall may still be perceived as unacceptable if its errors are novel, opaque, or difficult to predict.

A transitional technology

Driverless cars are not a finished product but a transitional one. They exist within a broader technological evolution in which incremental improvements coexist with unresolved risks.

Experts generally agree that, in the long term, autonomous vehicles are likely to reduce overall traffic fatalities.

However the present moment is one of uneven maturity. Some systems perform exceptionally well in structured environments, while others struggle with the unpredictability of real-world conditions. Regulatory frameworks are still evolving, as governments attempt to balance innovation with public safety.

Are driverless cars safe?

The answer is both yes and no.

They are safer than human drivers in many controlled and measurable respects – less prone to error, more consistent, and increasingly capable of avoiding common forms of collision.

But they are not yet safe in an absolute sense. Their failures, although rare, reveal fundamental limitations in how machines perceive and interpret the world. When those limitations are exposed, the consequences can be severe.

We stand therefore in an intermediate age – one in which the promise of automation is visible, but not yet fully realised. The road ahead is likely to be safer than the road behind. But it will not be without its accidents.