Tesla Accused of Giving European Regulators Misleading ‘Full Self-Driving’ Safety Data

A safety statistic can look decisive until its denominator, definitions and assumptions are examined. Tesla is now facing that problem in Europe, where the company has been accused of supplying regulators with misleading evidence to support approval of Full Self-Driving, or FSD. Correspondence obtained by Reuters shows Tesla shared self-published figures with authorities in the Netherlands and Sweden while arguing that broader use of the system would make roads safer.

The dispute does not establish that FSD is unsafe, nor does it erase more than 18 months of testing conducted by the Dutch regulator. Instead, it raises a narrower but crucial question: did Tesla present a fair comparison when it said vehicles using FSD could travel several times farther between crashes than ordinary human-driven vehicles? The answer could shape both European approval and public trust in increasingly capable driver-assistance systems.

The Allegation Reaches Beyond Advertising

Reuters reported that Tesla approached the Netherlands Vehicle Authority, known as RDW, in late 2024 to begin the European approval process. In a November letter, the company linked to its safety report and said increased use of FSD “leads to safer roads.” After the Dutch approval was announced in April 2026, a Tesla policy manager sent Swedish officials a presentation claiming cars using FSD travelled more than seven times farther between crashes than the average American vehicle. Independent safety researchers consulted by Reuters said the comparisons behind that message were invalid or distorted.

That distinction matters. The allegation is not simply that Tesla used enthusiastic language in a consumer advertisement. The information was circulated in a regulatory setting, where authorities must decide whether software capable of steering, braking, accelerating and turning should be permitted on public roads. Tesla did not respond to Reuters’ detailed requests for comment. RDW, however, said its decision did not depend on marketing claims or outside statistics and was based on its own testing, analysis, verification and auditing.

The Headline Numbers Sound Transformative

One Tesla presentation sent to Swedish regulators claimed that widespread use of FSD could potentially save 32,000 lives and prevent 1.9 million injuries. The same material relied on a claim that FSD-equipped Teslas could travel more than seven times farther between collisions than the average U.S. vehicle. Tesla leaders have gone further in public, at times describing the system as up to 10 times safer than human driving or pointing to an 85% reduction in crashes.

Those figures create an emotionally powerful picture: tens of thousands of families spared a death and millions avoiding injury. The problem, according to researchers who reviewed the calculation, is that the projection assumes every vehicle on American roads could effectively be replaced by an FSD-enabled Tesla and retain the claimed safety advantage. That imagined fleet includes motorcycles, freight trucks, old cars and vehicles used in conditions unlike those in which FSD is typically activated. A large projected benefit may be mathematically consistent with its assumptions while still being unrealistic as a forecast of what would happen on actual roads.

Not Every “Crash” Is Counted the Same Way

Tesla’s current methodology defines a collision using vehicle telemetry. It counts events involving deployment of an airbag or another non-reversible restraint, as well as lower-severity impacts that meet a specified change-in-velocity threshold. It also assigns a collision to FSD when the system was active at any point during the five seconds before the event. Tesla says this captures incidents in which a driver or the software disengages shortly before impact and avoids making subjective judgments about fault.

The controversial comparison arises when those internally detected events are placed beside federal crash estimates. Reuters’ experts said Tesla had compared FSD crashes involving airbag deployment with a broader national pool that included many less-serious, police-reported collisions. Tesla’s updated report now explains that it uses federal mileage totals and several NHTSA databases, selecting the Crash Investigation Sampling System for its “major collision” baseline because it focuses on crashes involving a towed passenger vehicle. Tesla also openly acknowledges unavoidable assumptions and differences in data collection. Even so, mixing thresholds, reporting systems or severity levels can produce a dramatic ratio that reflects methodology as much as technology.

Newer Cars and Easier Miles Can Tilt the Result

Vehicle age is another major source of distortion. Tesla argues that its pre-2014 vehicles without active safety features are a useful proxy for the average U.S. vehicle because the national fleet is roughly 12 years old. Critics counter that a modern Tesla is being compared with a much older mix of cars, trucks and motorcycles. Newer vehicles generally include stronger structures, automatic emergency braking, forward-collision warnings and other protections that can reduce crash risk regardless of whether FSD is operating.

Driving exposure creates a second complication. FSD users can decide when to switch the feature on, and Reuters reported that Tesla’s own data shows the system is used mostly on highways. Highways eliminate many intersections, pedestrians and crossing conflicts found on urban streets. A cautious owner may also disengage automation before a difficult construction zone, confusing junction or severe weather event. That selection effect can leave automated miles disproportionately concentrated in situations where crashes are less likely. A scientifically persuasive comparison therefore needs matching for road type, weather, geography, vehicle age, time of day and other conditions—not simply total miles divided by recorded collisions.

“Full Self-Driving” Still Requires a Human Driver

Despite its name, FSD Supervised is not legally or technically treated as a fully autonomous driving system in Europe. It can control steering and speed and perform complex manoeuvres, but the driver must watch the road, remain responsible and be prepared to intervene immediately. RDW says the European version monitors the driver’s eyes and availability to take over; repeated inattention can trigger warnings and temporarily prevent the system from being activated.

The branding has nevertheless troubled some regulators. Swedish officials discussed whether “Full Self-Driving” could give consumers a false impression of the system’s abilities, while Nordic authorities questioned its behaviour on icy roads, at higher speeds and around hazards such as moose. NHTSA also distinguishes Level 2 assistance from automated driving systems: Level 2 can provide steering and speed support, but the human must remain continuously engaged. That gap between capability and responsibility is central to the safety debate. A system that performs well for long stretches may encourage overconfidence precisely because the driver is still expected to rescue it during the rare moment it fails.

The Dutch Approval Was Based on Separate Testing

The Netherlands became the first EU country to grant provisional approval to FSD Supervised on April 10, 2026. RDW said it examined the European system for more than a year and a half on test tracks and public roads. The authority described it as driver-controlled assistance rather than a self-driving car and said correct use could make a positive contribution to road safety. It also stressed that Europe receives a different software version from the United States, making a direct one-to-one comparison inappropriate.

This is the strongest counterweight to the allegation about Tesla’s American statistics. RDW told Reuters that it did not rely on marketing claims or external figures, and that Tesla collected substantial test data which the regulator validated, tested and audited. At the same time, RDW has not publicly released the detailed research or datasets behind its decision. That leaves outsiders unable to independently reproduce the assessment or determine how the system performed across unusual conditions. The Dutch approval therefore demonstrates that FSD passed a lengthy regulatory process, but it does not by itself settle the separate dispute over the accuracy of Tesla’s public safety comparisons.

Europe’s Approval Process Is Still Unfolding

Tesla is using an exemption route under European vehicle law for new technologies not fully covered by existing rules. That route allowed RDW to issue a provisional approval valid in the Netherlands while seeking broader recognition. EU-wide authorization requires support through the relevant European process; Reuters reported that approval would need at least 15 of the 27 member states representing 65% of the bloc’s population. Until then, individual countries can choose to recognize the Dutch decision or issue their own permission.

The rollout has already expanded. By June 10, Belgium had become the fifth EU country to authorize the supervised software, after the Netherlands, Lithuania, Estonia and Denmark. Yet regulatory enthusiasm is not uniform. Swedish and Finnish officials have raised questions about speeding, winter roads and system naming, while safety advocates have asked for greater transparency and independent verification. The resulting patchwork means Europe is conducting two debates at once: whether this particular version of FSD meets technical requirements, and whether the evidence used to promote it is rigorous enough for a technology that could eventually reach millions of vehicles.

American Scrutiny Adds to the Pressure

European officials are evaluating FSD while the technology remains under active scrutiny in the United States. In October 2025, NHTSA opened an investigation covering an estimated 2.88 million Tesla vehicles after reports involving alleged traffic-law violations, including proceeding through red signals and travelling against the proper direction of traffic. The agency’s initial file identified 58 incidents from complaints, media reports and required manufacturer submissions. A separate investigation opened in 2024 examines FSD performance in reduced-visibility conditions such as glare, fog and airborne dust.

An investigation is not a finding that a defect exists, and raw incident totals cannot establish comparative risk. NHTSA itself warns that crash data from advanced driver-assistance systems are not normalized for fleet size, miles travelled or operating conditions. Manufacturers also differ greatly in what their vehicles can detect and transmit, so a highly connected fleet may report more events simply because it sees more of them. Those cautions cut both ways: Tesla’s telemetry may provide unusually broad information, but neither high incident counts nor impressive miles-per-crash figures should be treated as definitive without comparable exposure data and consistent definitions.

Credible Safety Proof Requires Comparable Evidence

A trustworthy safety case would begin with like-for-like comparisons. Researchers evaluating automated driving commonly match or adjust data for geography, road class, weather, lighting, traffic environment and mileage. A 2024 peer-reviewed study in Nature Communications used matched case-control methods to compare autonomous and human-driven crashes under similar circumstances. Another peer-reviewed analysis involving Swiss Re and Waymo calibrated its human benchmark by mileage and ZIP code and used insurance claims to assess bodily injury and property damage.

Tesla has a valuable starting point: a connected fleet capable of producing billions of telemetry packages and identifying events soon after they occur. The next step would be to give qualified independent researchers controlled access to sufficiently detailed, privacy-protected data, publish confidence intervals and explain every inclusion rule. Results should separate highways from urban streets, major impacts from minor contact, newer vehicles from older ones and supervised assistance from genuine driverless operation. Until that standard is met, the European dispute is likely to persist. The question is not whether automation can eventually save lives, but whether regulators and the public are being shown evidence strong enough to prove when it already does.