The NTSB's preliminary report on the Katy, Texas crash confirms that the Tesla Model 3 driver manually overrode Full Self-Driving by pressing the accelerator to 100 percent before the vehicle struck a home at speeds greater than 70 mph, killing 76-year-old Martha Avila. (The Verge)

The crash occurred in June on a two-lane road in Katy, Texas, with a posted speed limit of 30 mph. Weather was clear, the roadway was dry, and daylight conditions were present at the time. Michael Butler, 44, was driving the Model 3 when it left the roadway and collided with the home. Avila was inside the residence and was killed. (The Verge)

According to the NTSB report, data extracted from the vehicle shows the driver depressed the accelerator pedal fully, which disengaged the FSD system. Tesla AI head Ashok Elluswamy stated separately that the driver manually overrode self-driving by pressing the accelerator all the way to 100 percent. (The Verge)

Butler was arrested and is facing manslaughter charges. He told hospital providers after the crash that he remembered putting the car in self-driving mode and that he passed out. (The Verge)

Data recovered from Butler's phone included FSD-related Google searches, among them "Tesla fsd not aggressive enough 2026 model" and "tesla fsd too timid." (The Verge)

The National Highway Traffic Safety Administration is also investigating the crash. (The Verge)

The NTSB's findings place the immediate mechanical cause of the crash on a deliberate driver override rather than a failure of the FSD system itself. The accelerator-to-100-percent input is what disengaged Full Self-Driving, meaning the vehicle was operating under manual control at the moment it exceeded 70 mph on a 30 mph residential road. This is a critical distinction for the ongoing investigations and for the manslaughter case against Butler, because it situates the driver's actions, not autonomous software behavior, as the proximate cause.

That said, the phone search history opens a separate and uncomfortable line of inquiry. The queries "Tesla fsd not aggressive enough 2026 model" and "tesla fsd too timid" suggest a user frustrated by the system's conservative driving behavior, someone who may have been inclined to override FSD precisely because he found it too cautious. If a driver's response to a system designed for safety is to force it out of its operating envelope, the technology is functioning as intended while the human-machine interaction is producing the opposite result.

Butler's statement that he passed out after engaging FSD adds another dimension. Tesla's driver monitoring system is designed to detect inattention and disengage the system with escalating warnings, but a driver who is unconscious and simultaneously applying full throttle creates a scenario the system's guardrails were not built to prevent. The accelerator override is a feature, not a malfunction; Tesla has long allowed drivers to resume manual control at any time by pressing the pedal, on the assumption that a driver pressing the accelerator intends to drive. Whether that assumption holds when the driver may be unconscious is a question neither the NTSB preliminary report nor Elluswamy's statement addresses.

For engineers and safety professionals working on driver-assistance systems, the Katy crash is a case study in the gap between system design and user behavior. The FSD system appears to have been operating correctly and conservatively, as designed. The driver overrode it. The vehicle then traveled at more than twice the posted speed limit into a home. A 76-year-old woman who was inside that home is dead. The technology did not fail in any detectable way. The interaction between the technology and its user did.

The NTSB's report is preliminary, and both NTSB and NHTSA investigations remain ongoing. The manslaughter case will likely turn on whether a jury sees Butler's accelerator override and loss of consciousness as criminal negligence or as something else entirely. The regulatory investigations may turn on whether the current design of accelerator-based FSD disengagement, combined with the limits of driver monitoring, represents an acceptable safety posture for systems operating on public roads in residential areas at 70 mph.