Tesla Put Cybercabs on Austin Streets and NHTSA Answered in Hours
Paid rides started September 4 in a two-seat car with no steering wheel, no pedals, and no mirrors. The federal audit query into how Tesla self-certified that vehicle opened the same morning.
Tesla held an invitation-only Cybercab debut in Austin on September 3 and began paid public rides in limited parts of the city on September 4.
NHTSA opened an investigation Friday morning, hours after the first cars carried paying passengers.
The probe is not about a crash. There hasn't been one. It is an Audit Query into the process and technical data Tesla relied on to self-certify that a vehicle with no steering wheel, no brake pedal, no accelerator pedal, and no mirrors complies with Federal Motor Vehicle Safety Standards — and specifically, the basis on which Tesla determined that certain FMVSS provisions did not apply to it at all. It covers roughly 1,000 vehicles.
Self-certification is the whole story
The United States does not type-approve cars. Unlike the EU, there is no regulator that inspects a design and issues a permission slip before it reaches the road. A manufacturer certifies its own compliance with FMVSS, puts a label on the vehicle, and sells it. NHTSA's role is enforcement after the fact.
That system works cleanly when the standards contemplate the vehicle. Many FMVSS provisions do not merely permit a steering wheel and pedals — they are written around them. Requirements addressing steering-column displacement, brake-pedal force, mirror fields of view, and driver-facing controls assume a driver's seat with a human in it holding something.
A car with none of those parts forces a question the standards were not drafted to answer: is the requirement satisfied, or is it inapplicable? Tesla answered that question itself, which is exactly what the self-certification system tells it to do. NHTSA is now auditing the answer.
This is the first serious test of whether American vehicle regulation can absorb a purpose-built driverless vehicle without new rulemaking. Waymo and Zoox have run driverless services for years, but Waymo's fleet is built on conventionally-certified vehicles with the controls still present. Tesla removed them.
The fleet is smaller than the announcement
Tesla did not say how many Cybercabs are carrying passengers. Texas DMV registration data shows 45 Cybercabs registered in the state.
Forty-five. In one city, in a geofence, on a service that launched with a closed event that annoyed some of the fanbase by not being public.
For scale: Tesla's robotaxi program reports more than 380,000 unsupervised miles logged without a notable incident, and the Cybercab runs Hardware 4.0 with FSD 14.3.3, targeting SAE Level 4 inside the geofenced area. Those are real numbers and they are not nothing. They are also the numbers of a pilot, and the gap between a 45-vehicle pilot and the mass-market autonomy story Tesla's valuation encodes is the entire argument.
The fares tell you what stage this is
On launch day, a Tesla robotaxi-branded ride from Montopolis to ACL Live quoted at $19.58. Uber, which partners with Waymo in Austin, quoted $12.96 for an EV on the same route. A separate 4.7-mile Cybercab ride came in at $9.62.
Earlier robotaxi fares started at a fixed $4.20, later $6.90 — joke pricing, which is a reasonable thing to do when the product is a demonstration. The move to market-ish pricing that lands above Uber is more informative. It says Tesla is no longer subsidizing to fill seats, and it says the cost structure of a 45-vehicle fleet with remote support does not yet beat a human driver in a Model Y.
Autonomy's economic argument has always been that removing the driver removes 60-some percent of the cost of a ride. That argument arrives only at scale, when the fixed costs of the software, the remote operations center, the depot, and the regulatory overhead are spread across a large fleet. At 45 vehicles, every one of those costs lands on a small number of rides.
What the audit query can actually do
An Audit Query is not a recall and not a defect investigation. It is NHTSA asking a manufacturer to show its work. The realistic outcomes range from the agency accepting Tesla's reasoning, to requiring changes in how the vehicle is certified or equipped, to concluding the vehicles were not lawfully certified — which would be the serious case, because it would apply to the whole build rather than to a component.
The speed is the signal worth reading. NHTSA did not wait for an incident, a complaint, or a quarter of operating data. It opened the query the morning after the first paid ride, on a vehicle whose design has been public since 2024. The agency had clearly been waiting for deployment to give it standing.
Tesla says it plans to expand gradually to more vehicles and more locations. That expansion now runs alongside an open federal question about whether the vehicle should have been on the road in the first place — a question that gets harder to answer favorably as the fleet grows, because the remedy for an adverse finding scales with the number of cars.
The part that is genuinely new
Strip out the noise and something real happened this week. A production vehicle with no human controls of any kind carried paying members of the public on American public roads. That had not happened before.
Waymo proved driverless works. Tesla is testing whether the regulatory system will let anyone build a car for it — and doing so by shipping first and answering questions second, which is the strategy that has worked for Tesla roughly as often as it has blown up.
Forty-five cars and a federal audit query is not a robotaxi network. It is the opening move in the argument about whether one is legal to build.