On 2 September 2026, French Transport Minister Philippe Tabarot said two Tesla vehicles running Full Self-Driving (Supervised) would begin on-road trials, following what he called a "constructive exchange" with Elon Musk. According to Euronews, Paris had declined to authorise the system in July on safety grounds. Five European countries have authorised FSD Supervised since the start of the year.
Testing is the right call. The way this decision was reached is the weak point. A reversal announced on X after a call with the vendor's chief executive tells the public nothing about what data the trial will produce, who will see it, or where it will sit.
The strongest case for caution
The sceptics' argument deserves a fair statement. A supervised driving system is a data-hungry product. Cameras record roads, pedestrians and other drivers' plates, and the vehicle logs a driver's trajectories. If that data flows to servers outside the EU, French regulators lose visibility into both the safety record and the privacy footprint. The July refusal was made on safety grounds. Reversing it within two months, with no published criteria, invites the suspicion that access to the minister mattered more than evidence.
That suspicion is reasonable. The remedy, though, is a clear and narrow data regime for the trial. A blanket ban on the technology would not supply one.
What EU law already requires
Much of the framework exists. Regulation (EU) 2019/2144, the General Safety Regulation, requires new vehicles to carry an event data recorder. Per the text, the recorder captures data such as speed, braking and safety-system activation. It must not store information that identifies the individual vehicle or its owner, and it is designed as a closed loop that overwrites older data. Access is meant for national authorities, for accident research.
That design is a useful template. It shows that a regulator can obtain a decisive safety record without building a permanent surveillance archive.
On the privacy side, the CNIL, France's data protection authority, published a recommendation on connected-vehicle location data on 30 June 2026. It says consent is required for location data unless the data serves a service the user has expressly requested (Article 82 of the French Data Protection Act, which transposes the ePrivacy Directive). It also asks for data minimisation and limited storage periods. The CNIL's earlier compliance pack goes further and encourages processing that stays inside the vehicle wherever possible.
A proportionate data regime for the trial
France can turn these principles into a short trial protocol. Four elements would do most of the work.
- Log what safety analysis needs. Record disengagements, driver take-over events, system-state transitions and near-miss triggers, tied to timestamps and road segments. Continuous video and full location histories should not be retained by default.
- Share incident data with the regulator directly. The regulator should receive it in a standard format, not through a vendor summary. Independent analysis is the only way to test a company's claims about its own system.
- Keep raw sensor data in the vehicle or in the EU. Video of public roads captures third parties who never consented to anything. Local processing, on the CNIL's own preferred model, avoids the question of foreign access.
- Set retention limits and publish them. A time-boxed trial should have a fixed deletion date for raw data and a public summary of what was collected.
These conditions are cheap for a two-vehicle trial. They would also make the results credible when other member states, or French courts, examine them.
Why this matters beyond Tesla
The stakes rise with scale. On 18 September, Estonian ride-hailing firm Bolt and US carmaker Lucid announced plans to deploy at least 25,000 fully autonomous vehicles across European cities, with a target of 100,000 by 2035. Bolt's chief executive said autonomous driving in Europe requires data, software, vehicles and operations to work as one system built for European roads and regulation. If that is right, then the data rules set in small trials today will govern much larger fleets later.
There is a real risk in overcorrecting. The Electronic Frontier Foundation has argued that automated licence plate readers cannot be made safe by configuration, because the harm lies in the mass surveillance itself. That is a strong argument against fixed roadside networks whose purpose is tracking. A supervised vehicle's cameras serve a different purpose, which is driving. The policy question is retention and access, not whether the sensors exist. Rules that treat both alike would either ban useful safety technology or leave real surveillance unchecked.
Conclusion
France should keep the trial and publish its terms. The ministry should state which data the trial will log, what the regulator receives, where raw footage may be stored and when it is deleted. It should also commit to releasing aggregate results whatever they show. That approach respects innovation because it lets a promising technology prove itself on French roads. It respects speech and privacy because it limits collection to what safety evidence requires. It also answers the July objection with evidence rather than a phone call. Regulators elsewhere in Europe, watching five peer countries move ahead, will be judged on the same standard.