Nevada just approved up to 8,000 robotaxis. The permit governs how many cars — not the cameras inside them.

The document that approved thousands of new cameras onto the streets of Las Vegas does not use the word 'camera' once. On Thursday, the Nevada Transportation Authority voted unanimously to let three companies operate as many as 8,000 robotaxis in Clark County over the next year. The permits are written in the language of transportation: vehicle caps, service areas, competition with the local livery trade. Read the record of the hearing and you will find an argument about traffic. You will not find an argument about surveillance. That absence is the story.

Here is what the authority actually approved. Tesla may deploy up to 5,000 vehicles, Waymo up to 1,000, and Uber up to 1,000, the last operating through its partners Motional and the Amazon-owned Zoox, which holds a separate permit for another 100. Eight thousand autonomous vehicles, ceiling to ceiling. The companies were candid that the ceilings are aspirational; Tesla's Cybercab chief engineer, Eric Early, told the authority that 5,000 'has always been a ceiling for us' and that he did not expect to field that many within the year. Whether the real number turns out to be 8,000 or a fraction of it, the approval sets the terms. And the terms are about how many cars, not about what the cars see.

What a robotaxi is, in evidence

To understand what was approved, describe the vehicle without the word 'car.' A modern robotaxi is a sensor platform that happens to carry passengers. A Waymo vehicle carries, by the company's own account, 29 cameras, alongside lidar and radar, running whenever the vehicle is operating — and, as reporting this year established, sometimes when it is parked. Tesla's system takes the harder line: it navigates on cameras alone, no lidar, which means the fleet Nevada just authorized in the largest number is also the one that sees exclusively through video. Multiply that by thousands and distribute it across a city, and you have not approved a taxi service. You have approved a mobile, continuous, privately owned recording network that drives down every street a passenger pays it to.

That network is not hypothetical, and neither is the demand for what it captures. Since 2023, police departments in San Francisco and in Maricopa County, Arizona, have issued warrants and subpoenas to Waymo for footage from its vehicles. This is on the record. In April, the Los Angeles Police Department released video captured by a Waymo as part of a hit-and-run investigation. The footage a robotaxi collects in order to drive itself is, the moment it exists, evidence — discoverable, subpoenable, and useful to any investigator who knows to ask. The company did not set out to build a surveillance tool. It built a driving tool that is indistinguishable, in the hands of a prosecutor, from a surveillance tool.

One incident this summer showed how short the distance is between the two. On July 6, in San Mateo, California, a Waymo carrying two 15-year-olds — who, the company said, appeared to be drinking and firing toy guns inside the vehicle — was remotely disabled by Waymo, which then alerted the police. Sit with the sequence. A private company, watching a live feed from inside its own vehicle, immobilized the car with the passengers in it and summoned law enforcement. Whatever you make of the teenagers, the capability is the point: the operator can see inside, can stop the vehicle, and can place the call. That is not a taxi. A taxi driver can do the last part. Only a networked camera platform can do all three.

The company did not set out to build a surveillance tool. It built a driving tool that is indistinguishable, in the hands of a prosecutor, from a surveillance tool.

What the permit does not say

Now return to the Nevada permit, because the gap is precise and worth naming. The record of the proceeding, as reported, sets vehicle caps and service conditions. It does not, on the public record, set a retention limit on the footage these vehicles will capture. It does not specify the legal standard — warrant, subpoena, or a phone call — under which that footage can be handed to police. It does not say who at the companies may pull a live interior feed, or under what rule. The opposition that did show up came from the Livery Operators Association and local taxi companies, whose attorney argued about 'the oversaturation of the commercial transportation industry' and crowding in the area operators call the Golden Triangle. That is a real concern. It is also a concern about competition, not about the cameras, and it was the loudest objection in the room.

Retention is the variable that decides how much of this matters, and it is the one no one was asked to rule on. Consider a case from San Francisco, where police sought footage of a robbery and Waymo did not produce it — not out of resistance, but because by the time the warrant arrived the company was no longer storing the video. That cuts both ways, and honestly. Short retention protects passengers from becoming a permanent, searchable archive; it also means the record is governed by a company's internal policy rather than by law. Waymo says it resists unwarranted requests and evaluates them case by case, complying only when legally compelled. I take the company at its word about its current practice. The problem with practice is that it is not a permit condition. It can change with a policy memo, a new owner, or a request the company decides not to fight, and the passenger who rode across town last Tuesday has no way to know which.

Follow the incentive and you find the structure every surveillance market shares. The value of the footage is a byproduct the companies did not have to pay for: they collected it to teach cars to drive, and it turns out to be worth something to a second customer they never signed — law enforcement. Civil-liberties advocates have taken to calling the result a 'police-vehicle-data pipeline,' in which every mile a paying passenger travels becomes a potential frame in a criminal file. The passenger funds the collection. The company owns the archive. The police query it. At no point in that chain does the person in the back seat get a say, and at no point in Nevada's approval was that chain examined.

Arrival by accumulation

None of this requires anyone to have acted in bad faith, and I want to be careful about that, because the mechanism does not need villains. The engineers put 29 cameras on a Waymo to keep it from hitting a cyclist. Tesla went camera-only to cut cost and complexity. The police who subpoena the footage are working real cases. Each decision is individually defensible. Collectively they assemble a surveillance capacity that no one designed, no one voted for, and — in Nevada this week — no regulator was asked to weigh. That is how these systems always arrive. Not by design, but by accumulation: each link reasonable, the sum unaccountable.

What would change it is not complicated, and it belongs in the permit rather than in a privacy policy. A cap on how long footage is retained, written into the license. A clear legal standard — a warrant, not a phone call — for handing interior or exterior video to police, with the number of such requests published. A rule on who inside the company may watch a live feed, and when. Nevada had the leverage to attach every one of those conditions on Thursday, because a permit is exactly the moment a regulator holds the pen. Instead it approved how many cars each company may run and left what those cars record to the companies. The vehicles will be on the road within the year. The cameras were approved this week. The rules for the footage were not written at all, and the record shows no one asked for them.