Tesla Disputes Autopilot Claim in Fatal Katy, Texas Crash

A Tesla Model 3 crashed through the brick front of a Katy, Texas home on the evening of June 20, killing 76-year-old Martha Avila. The driver, Michael Butler, told Harris County sheriff's deputies that Autopilot had been engaged at the time. TechCrunch reported on Tesla's counter-narrative before investigators had completed their independent review.

Before any independent analysis of the vehicle's data logs, Ashok Elluswamy, Tesla's vice president of AI software, posted on X that "the driver manually overrode self-driving by pressing the accelerator all the way to 100% of the accel pedal in this residential area," adding that the vehicle "reached a speed of 73 mph during the crash, and had the accelerator pressed even after the crash." Elon Musk amplified the post, writing "FSD drives slowly through neighborhood streets and this was a high speed crash!" Both claims remain unverified by any party outside Tesla.

The National Highway Traffic Safety Administration opened a special crash investigation, one of more than three dozen such probes into Tesla crashes involving partially automated driving systems since 2016. This particular incident falls within a broader 3.2-million-vehicle NHTSA probe already at Engineering Analysis status, the final step before the agency can demand a recall.

The honest precedent to know: a 2021 Harris County crash initially attributed to FSD was later cleared by the National Transportation Safety Board, which found "no use of the Autopilot system at any time during this ownership period." That earlier incident also involved near-full accelerator application in a residential zone. Tesla's version of events may ultimately prove correct, and the vehicle's event data recorder will be the evidence that settles it. But the company's executives publicly stated their telemetry findings on X before investigators had independently reviewed that data.

What the reporting does not give you is whether Tesla's FSD software could or should have overridden the accelerator input, or what software version was running at the time. Those details are central to how supervised automation is designed and defended. NHTSA's engineering analysis will resolve less than either side's early public statement suggests.