Tesla has spent a decade promising that its cars will soon drive themselves.
This week, the company quietly released a fresh batch of numbers showing how often human drivers still have to grab the wheel.
The figures land squarely in the middle of a national argument about whether we are testing safety technology or selling a science experiment to people stuck in traffic.
According to the automaker's latest safety report, drivers using Autopilot disengaged the system roughly once every few hundred miles in certain conditions.
Tesla frames these as rare events, a sign the software handles the boring parts of commuting.
Critics counter that a single forced takeover on a rain-slicked highway at 70 mph is not a statistic.
It is a person yanking a steering wheel while their kids sit in the back seat.
The real trouble is what these numbers do not tell you.
Tesla reports disengagements, not near-misses.
A driver who taps the brake because a shadow crossed the lane looks identical in the data to a driver who avoided a ladder lying on the shoulder.
Owners describe a strange psychological trap.
The car handles the easy stretches so well that attention drifts, and the moment it needs you, you are already a half-second behind.
Safety researchers have a name for this: automation complacency.
It is a feature of human brains, and every automaker chasing hands-free driving is betting against it.
The stakes hit home in American daily life because these are not test vehicles on a closed track.
They are the same cars dropping teenagers at soccer practice and grandparents at church.
When a system fails, it fails on the roads we all share, in school zones, in construction chokepoints, in the rain on I-95.
Federal regulators have opened inquiries, but investigations move slower than software updates.
Meanwhile, the marketing keeps sprinting ahead of the engineering.
Names like "Full Self-Driving" suggest a capability that the fine print quietly walks back.
Drivers get a nag screen telling them to keep their hands on the wheel of a car they were told could drive itself.
That gap between the promise and the product is where trust goes to die.
None of this means the technology is worthless.
Adaptive cruise control and lane centering genuinely reduce fatigue on long hauls, and competitors offer similar features with humbler names.
The question is not whether machines can assist.
It is whether we have decided, as a society, that convenience is worth a body count we refuse to measure clearly.
What is missing is honesty at the point of sale and data the public can actually audit.
Every automaker collecting this information should publish disengagement and intervention rates in plain language, not press-release percentages.
Until then, drivers are unwitting test subjects in a nationwide trial with no consent form.
We keep outsourcing judgment to software and calling it progress.
Final Thoughts
A car that mostly drives itself is still a car that sometimes does not, and the difference is measured in lives, not miles.