By Rafael Patron, president of the Scientific-Technical Committee at AIPIA (the Italian Association of Artificial Intelligence Professionals)

Rafael Patron, president of the Scientific-Technical Committee, AIPIA

A van driver ends a long shift and finds a message waiting in the app: two harsh braking events, one possible phone distraction, a short coaching clip to watch. Whether that clip is opened with curiosity or with the feeling of having been caught has little to do with the camera, and a great deal to do with what the company told drivers before the system was switched on. That is the part of AI fleet safety that no supplier can install.

Safety pressure meets driver trust

International Fleet World’s report on its AI in the Cab webinar put the tension plainly. Fleet managers are being pushed by insurers, by senior management and by the Driver and Vehicle Standards Agency (DVSA) to bring down collisions and claims, and in-cab AI now offers automated driver coaching and smart daily briefs. The same discussion kept returning to one condition: drivers should not come away feeling watched or micromanaged.

That condition deserves more than a reassuring sentence in the launch email. A camera that drivers distrust gets covered, unplugged or argued over at every review, and the data it produces becomes a source of grievances instead of learning. A system that drivers understand, and that treats them fairly when it gets things wrong, tends to be accepted surprisingly quickly, because most professional drivers want to get home safely and have no wish to be the one in the incident report.

The difference is organisational. It sits in a handful of rules that a fleet can write down before go-live, and keep.

Tell drivers what the AI sees, and why

Start with a plain-language notice, separate from the contract and from the privacy policy that nobody reads. It should say which cameras are fitted, whether they face the road, the cab or both, and whether audio is captured. It should explain what makes the system save a clip, who can watch that clip, how long it is kept and when it is deleted. It should also say what the data will never be used for.

The Information Commissioner’s Office covers this ground in its guidance on monitoring workers. Employers are expected to make sure staff know what personal data is collected and how, to consider whether a less intrusive method would achieve the same aim, and to seek workers’ views before introducing new forms of monitoring. For a fleet, that last point is practical as well as legal: a conversation with a few experienced drivers before the pilot will reveal the situations where the system is likely to misfire, from mirror checks that look like distraction to bumpy depot entrances that register as harsh braking.

Minimisation matters here. If a road-facing camera with event-triggered recording meets the safety goal for a given vehicle type, a continuous cab-facing feed needs a stronger justification.

A person reviews before anything happens

An AI alert is a signal that something may have happened. It is not a finding. Detection models produce false positives, and the people who know the road, the vehicle and the schedule are better placed to judge context than a score on a dashboard.

So the second rule is simple to state and harder to keep under time pressure: no consequence for a driver follows from an alert until a trained person has watched the footage, considered the circumstances and recorded a decision. The driver gets the chance to explain before that decision is final. A harsh brake might be the best thing the driver did all week. The ICO’s guidance also flags that workers who ask for human intervention in automated decisions should not be disadvantaged for doing so, which is a useful line to quote in the fleet’s own policy.

Keep a short log of reviewed alerts, including the ones dismissed. Over a few months it shows where the system is unreliable and which thresholds need adjusting.

Keeping AI fleet safety on the coaching track

Coaching and discipline need separate tracks, with a written rule about when, if ever, a clip moves from one to the other. A sensible version reserves disciplinary use for a defined set of cases, such as a collision, deliberate dangerous driving or a repeated pattern after coaching has been tried, and requires a named manager to make that call. The algorithm never decides it.

Measurement follows the same logic. Each driver’s progress against their own baseline says far more than a public league table, which mostly teaches people to game the score. Some fleets also let drivers flag clips that show them handling a hazard well; it costs nothing and changes how the camera is talked about in the canteen.

Drivers’ representatives, where they exist, should see the coaching rules before launch and at each review.

What fleet managers need to know about AI

None of this works if the person reading the dashboard cannot explain it. Fleet managers do not need to become data scientists, but they do need to know roughly how a detection model reaches an alert, what it cannot see, why scores drift when routes or vehicles change, and how to explain a decision to a driver without hiding behind the software. The HSE’s leaflet Driving at work: managing work-related road safety is written for the people responsible for fleet management, among others. AI adds a layer they have to understand rather than simply trust.

For fleets that also run vehicles in the European Union, the EU AI Act, Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, is worth knowing. Annex III lists AI systems used to monitor and evaluate the performance and behaviour of people in work relationships among high-risk uses. Article 26 sets duties for the organisations that deploy them, including informing affected workers. The current Article 4 asks providers and deployers to take measures to develop the AI literacy of their staff. Which in-cab features fall into which category depends on what they actually do, and that analysis is worth doing early, not after a works council asks.

The Associazione Italiana Professionisti dell’Intelligenza Artificiale (AIPIA), the Italian association of AI professionals, works on this meeting point between the people who build AI systems and the organisations that deploy them. It publishes a guide to the EU AI Act that sets out risk categories, deployer obligations and AI literacy in practical terms. For a UK fleet manager with European operations, it is a quick way to see which questions to put to a supplier. The checklist below is also the kind of material that travels well into Fleet World’s webinars and Masterclasses, where fleet managers compare notes on what has actually worked with their drivers.

A short checklist before switching it on

• A one-page driver notice covering cameras, triggers, access, retention and the uses that are ruled out.
• A pilot with a small group of drivers, and their feedback written down.
• A trained reviewer for every alert, a right for the driver to explain, and a log of decisions.
• A written boundary between coaching and discipline, with a named decision-maker.
• Progress measured against each driver’s own baseline.
• A review date, three months after launch, to check false positives and thresholds.

Six months in, the most telling measure will probably not be on the dashboard. Ask a driver what the camera is for. If the answer is about getting home safely, the rules are doing their job.