From a CNN report:
A Tesla spokesperson released a moment by moment description of what happened in the 40 seconds before the crash.
After 15 seconds of what was described by Tesla as "visual warnings and audible tones," the autopilot began to disengage because the driver's hands were still not on the wheel.
About 25 seconds before the crash, "Autosteer began a graceful abort procedure in which the music is muted, the vehicle begins to slow and the driver is instructed both visually and audibly to place their hands on the wheel," according to the company.
Tesla said the driver responded 11 seconds before the crash by retaking the wheel, turning it toward the left and pressing on the accelerator.
"Over 10 seconds and approximately 300m later and while under manual steering control, the driver drifted out of the lane, collided with a barrier, overcorrected, crossed both lanes of the highway, struck a median barrier, and rolled the vehicle," according to Tesla's account.
So, the man never made the decision to disable autopilot. Instead, the car turned off the autopilot on its own. So, Musk could say with a straight face that autopilot was off. But how can a safety mechanism be allowed to turn itself off under any circumstances? Talk about the exact opposite of fail-safe.