The Sealed File: How a Supervised Driving System Crossed Europe’s Borders Before Europe Decided

by Francis Rozange | Sep 28, 2026 | 0 comments

On Tuesday 6 October, at a meeting convened by the European Commission’s directorate-general for the internal market, national type-approval officials will spend twenty-five minutes on the third item of a six-item agenda. The line reads, in full, “Continuation of discussions on request of the Netherlands for an Art. 39 authorisation.” It is scheduled between 11:45 and 12:10, after an update on the Euro 7 emission rules and an exchange of views on engines for non-road machinery, and before lunch. Nothing in the wording names the product at stake, and nothing in it announces a vote. Yet the item concerns one of the most closely watched driver assistance systems in the world, one that drivers in seven member states can already switch on today.

The system belongs to an American electric-car company and is sold under a name that promises complete self-driving before qualifying that promise in parentheses. Its path into Europe did not run through a new European rule. It ran through an exemption clause written in 2018 for technologies the rulebook had not anticipated, a provisional approval granted by the Netherlands, and a chain of national acceptances that followed without the underlying safety evidence ever being made public. For carmakers, insurers, fleet buyers and regulators, the story is less about whether machines can drive than about whether Europe’s system of mutual trust between approval authorities can carry a decision this large.

A Short Agenda Item With a Long History

The draft agenda of the 119th meeting of the Technical Committee on Motor Vehicles, dated 11 September and published in the Commission’s comitology register, confirms that a session the manufacturer had publicly pointed to, and the trade press had relayed, as the moment of a Union-wide vote is nothing of the kind. The Dutch request appears as a continuation of discussions. It had already been on the table at the committee’s meetings of 5 May and 30 June. No date for a vote had been announced publicly by 28 September. Reports in the specialist press say the committee will not reconvene before December, which would make that the earliest possible date for a decision.

The minutes of the May meeting explain part of the delay. According to the Commission’s summary record, the Dutch delegation “expressed a desire for a decision to be made by 30 June,” and the Commission’s representative answered that this timeline “would not be feasible due to the need for further data analysis.” The same record notes that the Commission asked the Netherlands, “for transparency reasons,” to present the file to two further expert groups. In other words, the central European institution in the procedure asked for more analysis and more openness before any vote. That request, rather than any single objection, frames what the October meeting is really about.

The Exemption Clause at the Heart of the Case

The legal route is Article 39 of Regulation (EU) 2018/858, the framework law for approving cars in the single market. The article allows a manufacturer to seek approval for a vehicle or system “that incorporates new technologies or new concepts that are incompatible with one or more regulatory acts.” Approval requires three things: an explanation of why the technology is incompatible with the rules, a description of its safety implications together with the measures taken “to ensure at least an equivalent level of safety,” and “test descriptions and results” proving that equivalence. The final authorisation belongs to the Commission, through an implementing act adopted after a committee vote.

While that decision is pending, the article lets the national authority grant a provisional approval “valid only in the territory of the Member State of that approval authority.” The authority must inform the Commission and the other member states “without delay by means of a file” containing the safety case. Then comes the clause that made the past six months possible: approval authorities of other member states “may accept the provisional EU type-approval” on their territory, provided they notify the issuing authority in writing. Nothing in the article obliges them to conduct their own tests. The mechanism was designed as a bridge for innovation, and it presumes a high level of confidence between regulators.

What the Dutch Authority Actually Approved

The Netherlands Vehicle Authority, the RDW, announced its decision on 10 April 2026. Its notice was unusually direct about what the system is not. A vehicle equipped with the system “is not self-driving,” the authority wrote, and it later added in capital letters that such vehicles “are NOT autonomous or self-driving.” The driver “remains responsible and must always remain in control.” Sensors check whether the driver’s eyes are on the road and whether the hands are available, although the hands “do not have to be on the steering wheel.” Persistent inattention triggers warnings and, eventually, a temporary lockout of the function.

The RDW stressed that the approval was the product of more than a year and a half of testing on its track and on public roads. It also pointed out that such approvals are not new in this field: an American carmaker already holds an Article 39 approval for hands-off driving on motorways, and a German manufacturer holds an approval for hands-off motorway driving combined with automated lane changes. What distinguishes the new case is scope. The Dutch notice itself acknowledges that the system “takes over multiple driving tasks” and supports the driver “more than other systems,” and the authority later said its own tests covered “complex and busy urban traffic” and not only divided highways.

Five Departures From the Rulebook

The exact content of the exemption is the part of the file the public can partly reconstruct, thanks to the manufacturer itself. In a web page it titles an “Article 39” dashboard, the company lists five requested departures from UN Regulation 171 in its 01 series. The system would perform manoeuvres it initiates itself and withhold requests for the driver to put hands on the wheel, first on highways and then on non-highway roads, “provided visual monitoring of the driver is possible.” Such manoeuvres “will not be inhibited based on the detected driver state” or on disengagement warnings. The system may exceed the lateral acceleration limits written into the regulation.

The fifth request is the one that has dominated the political debate. In the company’s words, “in certain cases” the system “will adjust the maximum speed to allow the system to drive the appropriate speed for the surrounding traffic conditions, even when this exceeds the detected speed limit.” These five points are company statements about a request, not a regulator’s summary of what was granted, and they should be read that way. They nevertheless make one thing plain. This is not a system that happens to sit at the edge of the existing rules; by the manufacturer’s own account, it departs from those rules on driver engagement, vehicle dynamics and speed.

A Regulation Written for Motorways, Stretched Toward Cities

UN Regulation 171 on driver control assistance systems was adopted by the World Forum for Harmonization of Vehicle Regulations, known as WP.29, in March 2024, and entered into force that September. According to the UNECE, it covers systems that assist “in controlling the longitudinal and lateral motion of the vehicle on a sustained basis, while not taking over the entire driving task,” equivalent to the level of automation engineers call Level 2. The driver “must therefore permanently monitor the surroundings.” A 01 series of amendments followed in 2025. The text has been evolving quickly, and the Dutch file sits inside that movement rather than outside it.

In June 2026 the Council of the European Union established through Council Decision (EU) 2026/1411 that the Union would vote in favour of a new 02 series of amendments at the World Forum’s session in Geneva, where it was adopted. The Commission’s proposal for that decision describes the change precisely: new requirements for system-initiated manoeuvres “on highways while withholding ‘hands on request'” and for such manoeuvres “in non-highway scenarios with ‘hands-on’.” The gap between that sentence and the Dutch request is the whole story in miniature. The international rulebook now contemplates city manoeuvres, but with the driver’s hands on the wheel. The exemption asks for them without.

Mutual Recognition Without a Common Reading

Since April, six member states have accepted the Dutch provisional approval. According to a tally published in the technology press on 8 September, Lithuania followed in May, Estonia in late May, Denmark in June and Belgium on 10 June, while Slovenia’s acceptance was announced by the manufacturer on 7 September. The Czech transport ministry announced its own acceptance on 21 September, saying it had joined “the Netherlands and five other European states.” In Belgium, the permission was granted by the Flemish mobility minister, Annick De Ridder, and applied to the whole country because, as the Dutch-language public broadcaster explained, an approval by one region “immediately” applies in the others.

By Eurostat’s count of population on 1 January 2026 (provisional for the Union total), these seven countries are home to about 53.4 million people, or 11.8 percent of the Union’s 452 million inhabitants. That share matters because it reveals how the procedure has been used. Article 39 was designed to let a pioneering technology operate in a limited territory while Europe reflects. It has instead produced a patchwork of acceptances by mostly smaller member states, while the largest markets, Germany, France, Italy, Spain and Poland, have not followed. Belgian drivers, a company spokesman told the broadcaster, are warned that the software will switch off as they approach a border beyond which the approval does not apply.

What the Other Capitals Saw, and What the Public Did Not

The phrase “sealed file” is accurate but needs precision. Article 39 requires the Dutch authority to send its safety file to the Commission and to the other member states, so national regulators are not voting blind. What remains sealed is the public version. In August, the RDW told the Dutch press that “test and technical information is considered commercially sensitive information and trade secrets,” and that information shared with the committee is confidential. The regulation supports that reading: Article 7 obliges approval authorities to “observe confidentiality in order to protect commercial secrets,” subject to its duty to make information available to the Commission and to other disclosure requirements in Union law.

The pressure behind that confidentiality came from the manufacturer. According to correspondence obtained by an international news agency under the Dutch open-government law and reported in August, the correspondence began at the end of 2024, and in April 2025 a company representative sought confirmation that a specific document would “never” be released and asked how the authority would guarantee it stayed out of public view. The RDW’s own statement of 17 June is candid about the consequence: manufacturers share commercially sensitive information, and “for that reason, we are not in all cases able to provide substantive information.” The public, the press, road safety researchers and insurers are therefore asked to accept a conclusion without seeing the evidence behind it.

The Arithmetic of a Qualified Majority

When a vote does come, the rules are demanding. Article 83 of the framework regulation sends Article 39 decisions to the committee under the examination procedure of Regulation (EU) No 182/2011, which requires the qualified majority defined in the Treaty on European Union. According to the Council, that means “at least 55% of member states vote in favour (15 out of 27)” and that those states “represent at least 65% of the total EU population.” An abstention, the Council notes, “counts as a vote against.” On 2026 figures, the favourable side must therefore gather roughly 294 million inhabitants, more than five times the population of the seven countries that have so far accepted the Dutch approval.

The other side of the ledger is the blocking minority, which “must include at least four member states” representing more than 35 percent of the population. Germany and France alone account for about a third of the Union’s inhabitants; add a third mid-sized state such as Sweden or Austria, and the total passes 35 percent, so that any fourth state would complete a blocking minority if all four voted against or abstained. There is a further subtlety. The framework regulation specifies that where the committee “delivers no opinion, the Commission shall not adopt the draft implementing act.” A divided committee does not default to approval. It leaves the provisional national approvals in place and the European decision suspended.

Why Paris Said Not Yet

France has stated its position in unusual detail, in a video answer to a citizen’s question, published with a transcript on 22 July by the transport minister, Philippe Tabarot, on the government’s public question platform. The minister wrote that “the system permits exceeding the speed limit,” and that a vehicle could drive “at 70 km per hour instead of 50” without an explicit request from the driver, “up to 50% above the authorised speed limit.” He added that the system “does not guarantee an optimal level of driver attention in town and during the most complex manoeuvres, such as lane changes, crossing intersections or roundabouts.” France concluded that “the safety trade-offs are not yet sufficient for an authorisation as it stands.”

The minister was careful to add that “our position is therefore not isolated,” a sentence addressed as much to Brussels as to the French public. The door did not stay closed. In early September, according to French technology press reports citing the minister’s own social media post, he announced after a conversation with the company’s chief executive that two equipped vehicles would be made available to the French authorities for road testing, which he called “a new stage.” Read together, the two statements describe a government that objects to specific functions rather than to the category of system, and that wants its own evidence rather than someone else’s file.

The Speed Offset and the Letter of the Law

Speed is where the regulatory and the political debates meet. According to the Commission’s summary record of the May committee meeting, “the concept of ‘speed offset’ was a subject of concern for some attendees, with two MS raising concerns that it could facilitate over speeding, contrary to existing regulatory frameworks,” while the Netherlands answered that it could help where posted limits are wrong and in adapting to traffic flow. A letter from the Swedish transport authority to the committee, dated 30 April and reported in June, warned that letting automated systems “systematically exceed legal speed limits” risks undermining the legal framework, and recommended a vote against unless the function was removed. The technology press has since reported that Sweden’s infrastructure minister said in early September that EU-wide approval should be put in place, so the letter may not settle Sweden’s vote.

Independent evidence on the point remains thin but is accumulating. A Belgian pedestrian and cyclist safety group drove the system for three days in July over roughly 400 kilometres in Belgium, and, according to a news agency report on 24 September, found that it exceeded the limit in most of the 30 km/h road segments it tested around Brussels, averaging 44 km/h there, and often displayed a 50 km/h limit despite recognising the 30 km/h signs. The group also reported that the system behaved cautiously around pedestrians. A spokesman for the Flemish government replied that the driver “remains 100% responsible.” The group said it had sent its report to the Flemish ministry and to the RDW, which did not respond to the news agency’s request for comment.

Berlin’s Preference for the Common Route

Germany, the largest vehicle market in the Union, has so far not acted alone. In a written reply published on 24 July on a German civic platform for questions to members of parliament, Christian Hirte, a Christian Democratic member of the Bundestag, stated that the Dutch decision is an approval under Article 39 and “an exemption approval,” not a Europe-wide authorisation. He said the approval documents were being evaluated with the Federal Motor Transport Authority, the KBA, from a road safety perspective, and that a test vehicle was to be provided to the authority in July so it could assess the system independently.

The German answer points to a second face of mutual recognition. Accepting a provisional approval is optional, and a country that declines is not rejecting the technology; it is reserving its judgment for the collective procedure the regulation prescribes. That posture carries weight in the committee arithmetic, because Germany’s population alone represents more than 18 percent of the Union. It also carries weight for the market. A system that cannot be used in Germany, France, Italy or Spain remains a regional feature, however many small states accept it, and every month of delay extends the period in which European buyers face different rules for the same software.

Two Vocabularies for One Product

For CSM International, whose content analysis work examines how products are described to the people who buy them, the most striking feature of the file is its vocabulary. The legal category is modest: a driver control assistance system, Level 2, with the driver fully responsible. The commercial name suggests something else, and even public statements drift between registers. A company spokesman told Belgium’s public broadcaster that the function is “an ordinary driving assistance system in the same category as cruise control.” The Czech minister, announcing acceptance of a Level 2 system, declared that his country “belongs among European leaders in autonomous driving technologies.”

The regulator’s own language has also moved. In April, the RDW wrote that thanks to continuous driver monitoring, “the system is safer than other driver assistance systems.” In its explanation of 17 June, published the day after the Dutch infrastructure minister explained the approval to the lower house of parliament, the same idea became “at least as safe as other driver assistance systems.” The change may be purely editorial, but it illustrates why wording matters in this field. Every public statement becomes a data point in how drivers calibrate their trust, and a single comparative adjective can carry more weight with consumers than a paragraph of legal qualification.

What Buyers Are Told, and What They Hear

UN Regulation 171 anticipated this problem. According to the UNECE, it requires manufacturers “to proactively communicate to users via all available means, including online, in advertising and at dealerships when purchasing a vehicle, about the limitations” of such systems and about the driver’s responsibility. That obligation turns marketing into a compliance question. When a product name implies full autonomy and a legal notice insists on the opposite, the question for regulators is not whether the fine print exists but which message the buyer retains. That is an empirical question, and it can be measured, which is precisely where customer research becomes useful to regulators and manufacturers alike.

Road safety researchers have long warned about the answer. In a September 2024 briefing on the proposed hands-off and system-initiated lane change provisions, the European Transport Safety Council cited research finding that drivers “were almost twice as likely to be driving distracted when the assistance systems were active,” and warned that allowing hands-off operation blurs the distinction between systems for which the driver is responsible and systems that take responsibility themselves. The word “supervised” carries the entire legal burden in the product name. Whether consumers read it as a condition of use or as a reassuring label is not something any published European study has yet established for this system.

The Evidence Europe Can See

What is public comes from two sources that should not be confused. The RDW states that its assessment involved more than 3,000 hours of testing, more than 1,000 test runs and data from 1.8 million kilometres driven in Europe with the system, and that its data experts “independently reviewed” the company’s statistical analysis. After approval, it raised the mandatory reporting frequency from annual to monthly. As of June, it said, nearly 40,000 vehicles with the system had driven about 24 million kilometres in the Netherlands “without any relevant incidents.” These are regulator statements, based partly on data the manufacturer supplies.

The company’s dashboard adds its own figures, including claimed reductions in emergency braking activations and in major collision rates against manual driving, drawn from North American fleet data, and a finding that where the system exceeded the detected speed limit in sampled European drives, it stayed at or below the median speed of surrounding traffic in 98 percent of high-speed cases. These are company claims, not audited results, and the dashboard itself concedes that the company “does not currently have the ability to directly stratify fleet-wide telemetry by all situational exposure variables,” such as weather, traffic density or lighting. Without the confidential file, outsiders cannot check how those limits were handled.

Liability Does Not Travel With the Software

For insurers, the approval changes less than the marketing might suggest, and more than drivers might assume. Under the Motor Insurance Directive of 2009, every member state must ensure that civil liability arising from the use of vehicles “is covered by insurance,” with the extent of liability determined by national law. Victims of a crash involving the system would be compensated in the first instance through the vehicle’s compulsory motor insurance, as they would be in any other car. Because every authority involved insists that the driver remains in control, the first line of liability stays, under national rules, with the driver or keeper, whatever the software was doing at the time.

The second line is changing. The new Product Liability Directive, which member states must transpose by 9 December 2026, explicitly defines a product to include software and treats a product as remaining “within the manufacturer’s control” when the maker can still push updates. In assessing defectiveness, courts must weigh “the presentation and the characteristics of the product,” its “reasonably foreseeable use,” and its ability “to continue to learn or acquire new features” after sale. For a system whose name, over-the-air updates and speed behaviour are all at issue, those criteria give insurers seeking recourse against a manufacturer a new vocabulary, and give marketing language a new legal weight.

What Insurers and Fleet Buyers Need to Know

In practical terms, a buyer or insurer confronting the system today faces three uncertainties, and they are worth stating plainly as analysis rather than prediction. The first is territorial: the function is lawful in seven countries and switches off at their borders, so cross-border fleets operate under two regimes on a single journey. The second is temporal: the approval is provisional, and if the Commission eventually refuses authorisation, Article 39 provides that the Dutch approval “shall be revoked six months after the date of the implementing act,” although vehicles built in conformity before then may still be registered in countries that accepted it.

The third uncertainty concerns information. Underwriters price risk on data, and the data that persuaded the Dutch authority are not available to them. Whatever positions insurers and their associations adopt, the confidential status of the file means that none can be based on the evidence the regulator saw. Fleet managers who allow the function should therefore document which software version runs in which country, how drivers are trained, and how monitoring alerts are recorded. None of that requires the sealed file, and all of it will matter if a claim, or a regulatory recall, ever tests the difference between assistance and delegation.

A Precedent Larger Than One Manufacturer

Whatever the committee eventually decides, the procedure will shape how other companies approach Europe. Article 40 of the framework regulation requires the Commission, once it authorises an Article 39 exemption, to “adapt the regulatory acts concerned” and, where a UN regulation is involved, to propose amendments in Geneva. The May minutes record exactly this concern, with one member state stressing the need for a plan to address the exemptions through future amendments to Regulation 171 and noting “the uncertainty surrounding the acceptance of these exemptions” in the UN working party. A European authorisation would thus become a lever on the global rulebook.

That is why competitors, not only regulators, are watching. A carmaker designing its own urban assistance system must decide whether to wait for the 02 series requirements, with hands on the wheel in cities, or to follow the exemption path and seek a provisional national approval of its own. For the industry’s competitive research teams, the practical lesson of the past six months is that a single national authority, followed by a handful of smaller states, can open a market in weeks, while the collective procedure moves in quarters. The asymmetry is built into the law, and it rewards companies that can persuade one regulator first.

A Test of Trust in the Single Market’s Machinery

The European Transport Safety Council argued in April that the committee “operates in private, with no formal route in for civil society, road safety experts, or anyone else outside the room,” and urged member states not to accept the Dutch approval nationally until there had been transparent deliberation and real-world evidence from Dutch roads. Six member states chose otherwise, each within its rights. The question their decisions leave open is not legal but institutional: how much weight one authority’s confidential judgment should carry across twenty-six other jurisdictions when the technology in question departs from the common rulebook on speed, attention and control.

The meeting of 6 October will not answer that question, and the absence of a vote is itself informative. It suggests that the Commission and a significant group of member states want more analysis, more independent testing and, perhaps, a narrower request before they commit the whole Union. The file that persuaded the Netherlands remains closed to the public, and the people who will ultimately live with the result, drivers, pedestrians, insurers and buyers, will have to judge it by the only material they can see: the behaviour of cars on the road and the words used to sell them.