Tesla's Belgian safety test found a 30 zone that looked like 50

01

WHAT ACTUALLY CHANGED

Reuters reported on September 24 that Belgian road-safety group Johanna.be had completed a three-day, roughly 400-kilometer test of Tesla Full Self-Driving (Supervised) in July.

The group tested the system on Belgian roads with particular attention to low-speed urban zones, pedestrians and cyclists.

Johanna.be found that the vehicle exceeded the limit in a majority of the tested 30-kilometer-per-hour segments around Brussels.

Across those tested 30 zones, the group reported an average vehicle speed of 44 kilometers per hour.

The group said the display appeared to recognize roadside 30 signs but often presented 50 kilometers per hour as the applicable limit.

Videos reviewed by the group showed the system attempting to overtake cyclists on streets where that maneuver was prohibited.

The report also credited the system with cautious behavior around pedestrians and cyclists, including proactive yielding.

Johanna.be sent its findings to the Flemish transport ministry and RDW, the Dutch regulator involved in the European approval process.

RDW approved the system in April 2026, and Belgium and some other European countries subsequently accepted that approval, according to Reuters.

An EU-wide vote on the system could take place on October 6. Reuters reported that several countries, including Sweden, had raised speed-limit concerns.

Tesla and RDW did not respond to Reuters' requests for comment on the Belgian test.

A Flemish transport ministry spokesperson said the driver remains fully responsible and must intervene when the system makes an error.

Tesla's public FSD page identifies the feature as supervised and says the currently enabled functions require active driver supervision.

Earlier in September, Tesla published European collision data that it says favors FSD-enabled driving. That dataset addresses collision frequency, not whether the system obeyed every local speed and overtaking rule.

02

WHY THIS MATTERS

A low-speed urban zone is designed around the vulnerability of people outside the car. Fourteen kilometers per hour above a 30 limit is not a cosmetic difference when stopping distance and impact energy rise with speed.

The apparent gap between recognizing a sign and adopting its limit points to a pipeline problem. Perception can be correct while localization, map data, rule selection, planning or display logic still fails.

Drivers do not supervise raw sensor output. They supervise a product that filters the road into a plan and an interface. A confident but incorrect display can weaken the oversight the system depends on.

Saying the driver is responsible establishes who must act. It does not show that the human received a timely, accurate warning or that the takeover was realistic in the moment.

Positive yielding behavior matters. Safety evidence should preserve the good results alongside the failures so the discussion does not become a highlight reel for either side.

The Belgian test is small and route-specific. It cannot estimate population-wide crash risk, compare all software versions or establish how often the behavior occurs across Europe.

Small tests can still expose testable failure modes. A repeatable mismatch between a detected sign and the chosen rule gives regulators a concrete scenario to reproduce rather than a vague argument about whether the system feels safe.

Collision-rate data and traffic-rule compliance answer different questions. A system could reduce some crashes while still speeding, making prohibited passes or confusing drivers about the applicable rule.

Tesla's earlier European comparison was company-produced. Reuters reported that underlying regulatory safety material was not publicly released because authorities treated it as commercially sensitive.

European approval can travel across borders. A decision made through one regulator can affect streets with different signs, road layouts, enforcement practices and local rules.

Software versions and map data change quickly. An approval tied only to a product name can become stale unless regulators know exactly which build, maps and behavior policy were tested.

Urban road rules are full of context: school hours, temporary works, zone entrances, road classes, painted restrictions and exceptions for cyclists. The hard problem is not reading a red circle. It is applying the right rule to the right place at the right time.

Supervision is a human-factors problem. Regulators should measure attention demands, time to understand an error and time available for correction, not merely whether a driver eventually touched the wheel.

If a system needs constant human correction for local rules, the useful metric is not only miles between collisions. It is how often the driver must catch a confident legal or behavioral mistake before it matters.

03

WHERE IT COULD HELP

• Build a regulator-owned route library containing 20 zones, 30 zones, school streets, temporary limits, construction signs, cyclist-priority streets and prohibited-overtaking segments.
• Repeat each route in daylight, darkness, rain, glare and partial sign occlusion using the exact production software and map version proposed for approval.
• Log the sign the perception system detected, its confidence, the map segment selected, the legal rule chosen, the planned speed and the value shown to the driver.
• Treat disagreement between sign recognition and the displayed limit as a first-class safety event even when the driver corrects the speed.
• Measure how long the mismatch persists and how much time the driver has before the vehicle reaches a pedestrian crossing, cyclist or zone boundary.
• Publish route-level results instead of only fleet averages so local authorities can see which street designs or signs create repeat failures.
• Record prohibited-pass attempts separately from completed maneuvers. A human intervention can prevent harm while still revealing a planning failure.
• Include pedestrians and cyclists in test design, but protect them with closed-course rehearsals and remote abort procedures before live-road trials.
• Compare FSD with attentive manual driving on the same routes, conditions and speed rules without pretending that one small comparison settles overall safety.
• Require the driver display to highlight when the camera sees a sign that conflicts with map data or the current behavior policy.
• Give local road authorities a rapid correction path for map and rule errors, with a visible audit trail and a deadline for retesting.
• Tie approval to named software, hardware and map versions, then require regression testing after a material update.
• Publish intervention categories, including speeding, illegal passing, right-of-way, lane choice, construction handling and driver confusion.
• Keep driver responsibility explicit while evaluating whether the interface supports that responsibility with accurate information and realistic reaction time.
• Separate evidence into rule compliance, collision avoidance, comfort and travel efficiency. A good score in one column should not erase a failure in another.