By Caitlin Taylor, Road Safety Manager, RoSPA
Artificial intelligence is already in most work vehicles, whether organisations have chosen it or not. Automatic emergency braking, lane-keeping support, speed limit warnings and cameras that spot a driver reaching for their phone now come as standard on many new vehicles, or arrive through an insurer’s telematics offer. For many fleets, AI hasn’t been a strategic decision. It has simply turned up.
That matters, because AI can make driving for work safer, but it can also introduce new risks that nobody has assessed. Through my work leading ScORSA, the Scottish Occupational Road Safety Alliance, I hear the same question from employers again and again: is this technology actually helping, and how would we know? The honest answer is that it depends less on the technology itself and more on how it is introduced, governed and explained to the people using it.
Not all AI in vehicles are the same
“AI” covers a wide range of systems in vehicles.
At one end are rule-based tools that follow fixed instructions, such as a warning triggered when a vehicle exceeds a mapped speed limit. They are predictable and easy to audit, but they cannot adapt.
In the middle are predictive systems that learn from data, such as models that forecast brake wear or pick up signs of drowsiness.
At the far end are adaptive systems that interpret their surroundings and make decisions in real time, such as route planning platforms that re-plan a driver’s day as conditions change.
The more autonomy a system has, the more oversight it needs. Training matters more, questions about accountability get harder, and errors become less visible. Every one of these systems can get things wrong.
Where assistive AI can reduce risk for drivers
Used well, AI targets some of the most persistent causes of work-related collisions:
- Driver assistance systems react faster than a person and don’t get tired. Intelligent speed adaptation helps drivers stay within limits, reducing both the likelihood and severity of collisions.
- Driver monitoring systems use in-cab cameras to spot early signs of fatigue or distraction, such as changes in blink rate or gaze, and alert the driver before a lapse becomes a crash.
Other systems work further upstream. Predictive maintenance picks up developing faults before they cause a failure on the road. Route optimisation can build realistic schedules that remove the time pressure that pushes drivers to speed or skip breaks. Collision avoidance systems linked to telematics create a record of near misses, showing where the next collision is most likely to happen.
There is some encouraging evidence from practice. In one insurer-led trial, RSA fitted AI-enabled dash cams to an HGV fleet, alerting drivers in real time to behaviours such as phone use or not wearing a seatbelt. Over 12 months, the fleet’s claims frequency fell from 29% to 8%. It was a single fleet, the results were reported by the company that ran the trial and there was no comparison group, so it shouldn’t be treated as proof. But it shows what is possible when risk is picked up at source.
Where AI in workplace vehicles goes wrong
The biggest risks of AI in cars and other vehicles are human, not technical.
Over-reliance. Drivers who confuse assistance with self-driving may stop monitoring the road properly. The message needs to be clear and repeated: these systems assist the driver, they do not replace them.
Alert fatigue. Frequent false alarms teach drivers to ignore warnings, including the ones that matter. Some will switch systems off altogether if they can.
Surveillance and trust. If monitoring data is used to catch people out, drivers will resist it, and fatigue will go unreported rather than being managed. Tiredness needs to be treated as a safety issue, not a disciplinary one.
Bias. Driver monitoring systems can produce more false alerts for people who wear glasses or head coverings, and for people with darker skin tones. Systems that read head and eye movement may also misinterpret neurodivergent or mobility-related behaviours as signs of risk. An alert that fires unfairly for some drivers is both a safety problem and an equality problem.
Pressure from the algorithm. Route optimisation built around delivery times can quietly trade safety for speed, leaving drivers feeling they have to “meet the machine”.
Treat AI like any other work equipment
None of this needs a new management system. AI in vehicles should go through the same risk assessment as any other workplace technology, including the hazards the AI itself introduces. A few practical steps make a big difference.
- Give someone ownership. Appoint a named person, ideally bridging health and safety and operations, to oversee how systems are used and to handle driver feedback.
- Complete a data protection impact assessment. Be clear about what data is collected, who can see it, how long it is kept and how drivers can challenge it.
- Involve drivers early. Consult staff and their representatives before rollout, and explain in plain language what each system does and doesn’t monitor.
- Ask suppliers hard questions. How was the system tested, and on whom? What are its known limitations? Can its outputs be explained and challenged? Has it been tested for bias across different drivers and conditions?
- Build safety into the settings. Set intelligent speed adaptation to on by default, with clear rules for temporary override, and make sure route planning respects rest breaks.
- Pilot before you scale. Run a short trial, measure before and after, and be prepared to adjust or pause.
- Track the right measures. Near-miss reporting should rise as trust grows. Collision rates per million miles and harsh driving events should fall.
Inclusion should run through all of this. Systems that offer visual, audio and haptic alerts, and allow settings to be adjusted, work for more people. Practical demonstrations help staff with lower digital confidence far more than a manual does.
To help employers work through these questions, ScORSA has published a free guide to AI in workplace transport, developed with Anthrometric. It covers the main types of AI, their risks and safer practice, along with procurement questions, KPIs and quick-start steps for fleet managers. Its starting principle is simple: AI can support human accountability, but it can never replace it.
Keep people at the centre
AI will keep changing, and quickly. What won’t change is the employer’s duty to manage the risks of driving for work. The organisations that get the most from this technology will be the ones that introduce it with their drivers rather than to them, question it as rigorously as any other control, and keep asking whether it is actually making journeys safer.
About the Author: Caitlin Taylor
Caitlin Taylor is Road Safety Manager at RoSPA and leads ScORSA, the Scottish Occupational Road Safety Alliance, a Transport Scotland-funded programme supporting more than 1,200 member organisations to manage occupational road risk.
References
IEEE Access (2020) Investigating bias in facial analysis systems: A systematic review. Available at: https://doi.org/10.1109/ACCESS.2020.3006051
RSA Insurance (2024) Reducing risks on the road: Introducing AI dash cams. Available at: https://www.rsainsurance.co.uk/news/case-studies/reducing-risks-on-the-road/
ScORSA and Anthrometric (2025) Artificial Intelligence in Workplace Transport: Practical Guidance for Safe and Efficient AI Adoption in Workplace Transport. Available at: https://share.google/itwabYvceVnXQt09X
