A Framework of Trust That Holds as Systems Learn

On July 22, 2026, the Minister of Transport set both a standard and a goal. This technical note supports both. It describes the assurance capability that keeps that framework true after a system goes into service.

France assesses a system at a moment in time. A driving system built on machine learning continues to change after that moment, because it updates its model. The Minister's own two reservations illustrate why continuity matters.

The contextual-speed behavior he described, where the vehicle can drive 70 km/h instead of 50 ("peut rouler à 70 km/h au lieu de 50") without an explicit driver request, and the concern that the system doesn't guarantee an optimal level of driver attention (“ne permet pas de garantir un niveau d'attention optimal du conducteur”), are both behaviors a later software version can reintroduce. An approval given today should stay valid tomorrow. That requires a mechanism for the evaluation to continue.

The European framework already assigns responsibilities here. UN R156 is primarily manufacturer-led on post-approval verification, and UN R171 adds an explicit manufacturer in-service monitoring and reporting obligation for DCAS in operation. The monitoring duty therefore rests with the party being monitored. France can strengthen the framework of trust by adding the independent layer above it. Three components do that work.

1. A conditional recognition that ties approval to defined verification obligations.
2. Change-significance criteria that state when an update to a learned driving policy is large enough to require re-evaluation.
3. Independent verification of the manufacturer's in-service monitoring, with assured reporting to the approving authority, so a recognizing state can see when an approved behavior has moved.

The country that defines this layer sets the reference the rest of Europe adopts. France raised the standard first. France is positioned to write how the standard is kept.

Fusion Collective offers these components as an independent technical contribution, available to the relevant services and to the autumn ecosystem work.