Tesla cabin display showing road visualization while Full Self-Driving Supervised is active

news July 22, 2026

NTSB Data Shows the Driver Overrode FSD Before Fatal Tesla Crash

Recovered Tesla data shows the driver commanded full acceleration before a fatal Katy crash. Here is what the NTSB finding proves—and what it does not.

By Marcus Holloway

The first federal evidence from a fatal Tesla crash in Katy, Texas, makes one point unusually clear: Full Self-Driving (Supervised) was not commanding the car’s acceleration in the final moments before impact.

In a preliminary report released July 15, the National Transportation Safety Board said electronic data recovered from the 2025 Model 3 showed that its driver manually overrode FSD by pressing the accelerator pedal to 100 percent. The car was traveling faster than 70 mph when it struck a home on June 19, fatally injuring an occupant inside. The driver sustained minor injuries.

That finding matters because early accounts of the crash focused heavily on the fact that FSD had been engaged. It also supports Tesla’s earlier statement that the driver, not the system, commanded the extreme acceleration.

But a preliminary report is not a final probable-cause determination, and one crash cannot prove that FSD is broadly safe or unsafe. The responsible reading is narrower: recovered vehicle data undercuts the claim that FSD itself suddenly accelerated the car in this specific event.

What the NTSB Found

The crash happened shortly after 8 p.m. on Rose Hollow Lane, a two-lane residential road with a 30 mph speed limit. The road ends at Blooming Park Lane, where through traffic must turn right. Security-camera footage showed the Model 3 continue straight through the intersection, leave the roadway and hit the residence.

The NTSB said the driver had engaged Tesla’s FSD (Supervised) system before the crash. Data recovered from the vehicle then showed the driver pressing the accelerator to its full position, manually overriding FSD’s acceleration control. The Model 3 exceeded 70 mph before impact.

Weather, visibility and road-surface conditions do not appear to offer an obvious explanation. According to the NTSB, the weather was clear, the road was dry and the crash occurred in daylight.

Those are preliminary facts, not a completed causal chain. The NTSB says all aspects of the crash remain under investigation, while the National Highway Traffic Safety Administration has opened a separate special crash investigation.

Pressing the Accelerator Is a Deliberate FSD Override

Tesla’s owner documentation explains why the pedal position is important. The Model 3 owner’s manual says a driver can press the accelerator to increase speed while FSD (Supervised) is active. It warns that doing so may compromise the system’s ability to maintain a safe following distance, respond to obstacles, obey traffic controls or avoid excessive speed.

In other words, FSD does not lock the human out of the controls. It is designed so the driver can supersede its acceleration command. That is consistent with the system’s legal and technical status as SAE Level 2 driver assistance, not an autonomous chauffeur.

The distinction is especially important here. Saying “FSD was engaged” describes the state of the car earlier in the event. It does not establish that FSD issued the acceleration command that produced the crash. The recovered data indicates the opposite.

It would also be too broad to say the preliminary report has completely “cleared” Tesla. The NTSB has not yet published a probable cause, evaluated every relevant human-machine interaction or issued safety recommendations. What it has done is resolve the central question about who commanded full acceleration in the final moments.

Why This Does Not End the FSD Safety Debate

Tesla can reasonably point to the Katy data as evidence that FSD was blamed too quickly in this case. The company cannot reasonably treat it as proof that other FSD safety concerns are unfounded.

NHTSA has a separate open investigation into alleged traffic-law violations while FSD is engaged. The agency’s October 2025 opening document identified reports involving red-light violations, travel into opposing lanes and incorrect movement through turn-only or through lanes. It cited 14 crashes or fires, 10 injury incidents and 23 reported injuries in the initial population under review. Those figures describe reports under investigation, not final defect findings.

Another NHTSA investigation is examining FSD performance in reduced visibility after crashes involving sun glare, fog or airborne dust. The agency’s opening summary said one of four triggering crashes involved a pedestrian fatality and another involved an injury. Again, the investigation is not itself a conclusion that FSD has a defect.

These cases ask different questions from the Katy crash. A system-commanded move through a red light is not equivalent to a driver pushing the accelerator to the floor. Treating every incident involving a Tesla as one continuous argument produces more heat than useful safety analysis.

The Human-Machine Boundary Is the Real Story

FSD (Supervised) occupies an uncomfortable middle ground. The software can steer, brake, accelerate, change lanes and navigate city streets for long stretches. Yet the person in the driver’s seat remains responsible for monitoring it and can override it instantly.

That creates two recurring mistakes in public discussion. One is blaming the software whenever a crash involves a car that had FSD enabled, even when vehicle data shows a human command. The other is assuming that because FSD handles most of a trip, the driver can stop supervising. Tesla’s own FSD support page explicitly says the system requires active supervision and does not make the vehicle autonomous.

The Katy investigation illustrates why electronic evidence is essential. Initial statements, product branding and viral video cannot reliably establish whether software or a person controlled the car at a critical moment. Pedal position, system state, vehicle speed and event timing can.

It also shows why language matters. “Self-driving crash” is an attention-grabbing label, but it can collapse several different scenarios into one phrase: a system may be controlling the car, a driver may have just disengaged it, or a driver may be overriding one part of the system while other assistance remains active.

What Happens Next

The NTSB will continue gathering evidence before determining probable cause. Its final work may address driver actions, system design, warnings, the way pedal overrides are communicated, or other factors that the preliminary update does not settle. NHTSA’s special crash investigation is proceeding separately.

For now, the evidence supports a precise conclusion. FSD had been engaged, but the driver commanded full acceleration before the Model 3 left the road and struck the home. Tesla was right about that part of the event.

Precision should cut both ways. This crash should not be used as evidence that FSD caused unintended acceleration when the vehicle data says otherwise. It also should not be used to dismiss unrelated, still-open federal investigations into situations where FSD itself allegedly commanded an unsafe maneuver.

That is less satisfying than a sweeping verdict for either side. It is also the only conclusion the available evidence supports.