A new Belgian road-safety test is raising questions about how Tesla’s Full Self-Driving (Supervised) handles some of Europe’s most tightly controlled urban streets. The nonprofit johanna.be found that FSD exceeded the legal 30-km/h limit in 16 of 29 Brussels test segments, or 55%, while frequently showing the supervising driver a higher speed limit than the road legally allowed.
The findings are significant because Brussels has operated with a general 30-km/h urban limit since 2021, with higher speeds permitted only on designated roads. They also arrive as European regulators continue considering broader authorization of Tesla’s technology. FSD Supervised can control steering and speed, but it remains a driver-assistance system: the person behind the wheel must remain attentive, responsible and ready to intervene.
The Headline Result Came From 16 of 29 Brussels Segments
Johanna.be systematically examined 29 road segments in the Brussels-Capital Region where the legal limit was 30 km/h. Tesla FSD exceeded that limit in 16 of them, producing the 55% figure highlighted in the report. Only three of the 29 segments combined a correct 30-km/h speed-limit display with vehicle speeds that remained at or below the legal limit. Another 10 segments stayed within 30 km/h despite the vehicle displaying an incorrectly high limit, sometimes because surrounding traffic prevented the Tesla from travelling faster.
There is an important detail behind the study’s reported 44-km/h average. Researchers defined “segment speed” as the maximum speed FSD reached while travelling through a particular segment. Among the 16 segments where speeding occurred, those maximum segment speeds averaged 44 km/h. That is different from saying the Tesla averaged 44 km/h throughout the entire length of every street. In 17 clearly signposted 30-km/h segments tested under free-flow conditions, however, FSD exceeded the limit in 10, or 59%, with the maximum speeds reached in those segments averaging nearly 46 km/h.
Researchers Tried to Remove Driver-Selected Speeding From the Equation
The testing used a recently registered Tesla Model 3 running software version 2026.21.100 and FSD version 14.2.2.6. Researchers configured the vehicle in standard mode, set its speed offset to zero and disabled the “contextual max speed” feature. Those choices matter because both settings can otherwise allow FSD to select speeds above the limit it believes applies. Testing took place from July 20 to July 22, 2026, in dry conditions, with a camera recording both the road ahead and the Tesla display.
The Brussels investigation involved 59 individual drives through the 29 segments, with some locations repeated as many as five times. Results were reported by segment rather than by individual run. The locations were chosen using Brussels Mobility mapping to identify places where roads permitting 50 km/h transitioned into 30-km/h sections, many around schools. That design also defines the study’s limits. It did not attempt to measure FSD performance across every type of Brussels street, and the city’s wider residential 30-km/h network was not systematically sampled. It was a focused road test using one vehicle and one software version, rather than a large peer-reviewed fleet study.
In Many Cases, the Tesla Appeared to See the Sign but Still Showed 50 km/h
One of the most striking findings involves the difference between what the Tesla apparently observed and what it subsequently told its supervising driver. Of 21 Brussels segments marked with F4a 30-km/h zone signs, the system visibly identified the 30-km/h restriction in 15 cases. Yet the permanent speed-limit display changed to 30 km/h in only three of those 15 cases. In 12, the display instead indicated 50 km/h. Across all 29 Brussels segments, the report calculated that the displayed limit was too high in about 90% of cases.
That distinction makes the result more complicated than a simple inability to see road signs. In some recordings, according to the researchers, a 30-km/h value briefly appeared before the system reverted to 50 km/h. The study also tested a conventional Citroën C3 traffic-sign-recognition system on seven comparable routes. The Citroën correctly changed its displayed limit to 30 km/h on all seven, while the Tesla correctly reflected the zone limit in only one of the comparable cases. That small comparison cannot establish how every competing system would perform, but it gave researchers evidence that sign visibility alone did not explain the Tesla results.
Why the Difference Between 30 and 44 km/h Matters
Brussels did not make 30 km/h the normal urban limit simply to reduce speeding tickets. Since January 2021, 30 km/h has been the default across the Brussels-Capital Region, except on specifically marked major roads where 50 or 70 km/h remains permitted. The policy is intended to improve safety, calm traffic and reduce noise, particularly in environments where cars routinely mix with pedestrians, cyclists and other vulnerable road users.
Speed has an outsized effect on both crash probability and injury severity. The World Health Organization says a 1% rise in average vehicle speed is associated with roughly a 4% increase in fatal-crash risk, while the chance that a pedestrian will die after being struck rises sharply as impact speed increases. Johanna.be used an older WHO speed-risk curve to estimate that a pedestrian struck at 44 km/h faced almost six times the fatality risk associated with a 30-km/h impact. Exact injury probabilities vary among studies and crash circumstances, but the broader safety relationship is well established: modest-looking increases in urban vehicle speed can produce disproportionately larger consequences when a collision occurs.
Problems Were Also Recorded in Home Zones and Cycle Streets
The Brussels 30-km/h finding was not the only concern identified. Researchers separately drove through 12 Belgian home zones, where the legal limit is 20 km/h and pedestrians may use the roadway more freely. FSD exceeded 20 km/h in five locations, or 42%, with the study’s maximum segment speeds averaging about 26 km/h when speeding occurred. Eight of the 12 home-zone tests also showed a speed limit on the Tesla display that was higher than the legal one. Because these locations were selected through convenience sampling rather than systematic sampling, the percentages should not be treated as national failure rates.
Six cycle streets were also examined. Belgian rules cap vehicles at 30 km/h in these zones and prohibit motor vehicles from overtaking cyclists. The Tesla displayed the correct 30-km/h limit in none of the six locations, according to the study. More troublingly, FSD attempted to pass cyclists multiple times during two tests, including four attempts on one approximately 600-metre street. The supervising driver aborted the manoeuvres before the Tesla reached the cyclists. Those observations highlight a challenge that goes beyond recognizing signs: driver-assistance software must correctly interpret locally specific traffic rules, which can differ even between neighbouring European countries.
The Test Also Documented Several Things FSD Did Well
The report was not uniformly negative. Researchers said FSD generally behaved carefully around pedestrians and cyclists outside the cycle-street overtaking problem. The vehicle reportedly yielded reliably at zebra crossings and sometimes reacted early to people approaching the roadway. In one example, it waited behind two joggers travelling in the carriageway rather than immediately attempting to pass. In another, it slowed for a cat that appeared ready to cross before the animal stopped on the pavement.
That context matters because performance of an advanced driving system cannot be reduced to one statistic. FSD can display sophisticated behaviour in difficult interactions while still making errors involving basic legal constraints. The Belgian results therefore do not demonstrate that FSD is broadly unsafe in every circumstance, just as a successful pedestrian yield does not establish that its speed management is reliable. The study’s strongest evidence concerns the particular environments it intentionally examined: 30-km/h zones, home zones and cycle streets. Its authors concluded that the system’s courteous interactions with vulnerable road users were genuine strengths, but argued that they did not erase repeated failures involving speed limits and local cycling rules.
Tesla’s Own European Safety Data Presents a Different Picture
Tesla has published a much broader evidence package supporting FSD Supervised. Its European engineering-fleet data says more than 793,000 miles were driven across eight countries without a major or minor collision attributed to FSD performance. Tesla also reports more than 230,000 individual scenario tests on fixed routes in six European cities, with an overall pass rate above 99% and no events it classified as safety-critical. Those results cover a much larger volume of driving than the Belgian nonprofit’s test.
Tesla has separately analyzed 708 European driving samples in which FSD travelled above its system-determined limit. After excluding low-speed scenarios, 680 samples remained; Tesla reported that the vehicle travelled at or below the median speed of surrounding traffic in 98% of those cases. The two bodies of evidence are not directly comparable. Tesla’s analysis specifically examined situations involving surrounding traffic and excluded low-speed cases below 25 km/h, while johanna.be deliberately concentrated on low-speed streets where the system sometimes appeared to determine the legal limit incorrectly in the first place. The contrasting results illustrate why sampling criteria can dramatically affect conclusions about a driving system’s performance.
The Study Lands During a Crucial European Regulatory Debate
FSD Supervised received provisional approval from Dutch vehicle authority RDW in April 2026 after an assessment lasting more than a year and a half. RDW stresses that the system is not autonomous: drivers remain responsible, must monitor the road and must be capable of taking over immediately. Under the European Union’s Article 39 process, other member states can individually recognize a provisional approval while the European Commission considers whether to authorize it more broadly.
Johanna.be said it sent its report to RDW and the relevant Flemish authorities on September 7 and asked them to reassess the approval. Reuters reported that Tesla and RDW did not respond to its requests for comment on the new findings. A Flemish transport ministry spokesperson emphasized that the driver remains fully responsible and said evidence supplied by Tesla indicates significant accident-prevention potential. European safety groups have meanwhile pushed for greater independent scrutiny of advanced Level 2 systems, particularly where increasingly capable software still depends on continuous human supervision. With broader EU authorization under consideration, the Belgian test adds a narrowly focused but timely piece of evidence to that discussion.
