On a Friday evening in Katy, Texas, a Tesla Model 3 left a residential street near Houston at high speed and struck a home, killing 76-year-old Martha Avila, who had been inside. She was airlifted to a hospital and later pronounced dead, according to the Harris County Sheriff’s Office. Now two federal agencies — the National Transportation Safety Board (NTSB) and the National Highway Traffic Safety Administration (NHTSA) — are investigating whether Tesla’s Full Self-Driving software played a role, according to CNBC and the New York Times. Tesla confirms FSD was active. The central question: who or what was actually driving?
What the Data Shows — and What’s Still Contested
Tesla’s own numbers paint a damning picture, but investigators want to verify every byte themselves.
Tesla’s Head of AI, Ashok Elluswamy, stated on X that vehicle data show driver Michael Butler “manually overrode” FSD by pressing the accelerator to 100%, reaching 73 mph before impact, as reported by Electrek. Elon Musk initially suggested the speed itself proved the car wasn’t “driving itself.” Then Tesla confirmed FSD was engaged when the vehicle departed the road. That contradiction is now central to the federal investigation.
NTSB and NHTSA will pull the event data recorder and Tesla’s onboard logs to independently verify pedal inputs, system warnings, and the exact sequence of driver and system actions. Independent verification matters here. In a 2019 Key Largo, Florida crash, a hacker recovered “collision snapshot” data that Tesla had told a court didn’t exist, according to the Washington Post. A federal jury awarded $243 million. That precedent explains why investigators aren’t simply taking Tesla’s public statements at face value.
Here’s where every investigation currently stands:
- Martha Avila, 76, was killed inside her home when the Tesla struck it
- Driver Michael Butler survived, cooperated with police, and reported using Tesla’s automated features
- No mechanical malfunction or driver intoxication found by local investigators
- Both NHTSA and NTSB will seek Tesla’s raw vehicle logs
- The Avila family has filed a wrongful-death lawsuit against Butler and Tesla, alleging negligence in driving and in FSD design and marketing
- No criminal charges filed yet; evidence forwarded to the Harris County district attorney
“Any claim by Tesla based on crash rate data must be considered suspect until independent review is completed.” — Phil Koopman, safety engineer (Substack).
For the estimated 3.2 million U.S. Tesla owners currently under NHTSA’s broader Autopilot engineering analysis — the final step before a potential recall — this crash adds pressure to an already tense regulatory environment. FSD remains a Level 2 driver-assist system. The human is legally responsible at all times, regardless of what the name suggests.
A Pattern Regulators Can No Longer Ignore
Dozens of federal probes and at least 65 reported deaths form the sobering backdrop to this investigation.
Spanning nearly a decade, NHTSA has opened more than three dozen special investigations into Tesla crashes involving Autopilot or FSD, according to CNBC. A crowdsourced database documents at least 65 deaths where these systems were reportedly contributing factors. The NTSB has cleared Tesla’s systems in some prior Texas crashes — the agency follows data, not assumptions — but the pattern of incidents keeps expanding.
Calling software “Full Self-Driving” while requiring constant human supervision is a bit like labeling a sous vide machine a “personal chef” — technically impressive, fundamentally misleading, and genuinely dangerous when someone takes the branding literally on a suburban street.
The road ahead could bring mandatory FSD speed restrictions in residential zones, stronger driver-monitoring requirements, or a broader regulatory push to police how automakers name and market driver-assist systems. Whatever the outcome, accountability runs through those data logs. Waymo Robotaxis represent a contrasting approach to autonomous vehicle deployment that regulators are also watching closely.
