A Waymo robotaxi zips down the wrong side of a downtown Austin street, cutting ahead of traffic. A police officer sprints toward it, then hits a wall no traffic academy prepared him for. There is no driver to pull over, no window to knock on, no license to hand across. When a computer breaks the law, who gets the citation?
That question is now playing out in cities across America, and the honest answer is that nobody has fully figured it out. As autonomous taxis from Alphabet’s Waymo, Tesla, and Amazon’s Zoox multiply on public roads, police departments are improvising an entire enforcement playbook in real time. The cars are often better behaved than the humans around them. The problem is what happens when they are not.
The paradox no traffic cop was trained for
Corporal Patrick Oborski of the Austin Police Department has learned this the hard way, according to reporting by The Wall Street Journal. He has watched robotaxis drift onto the wrong side of the road and, on one occasion during a University of Texas football game, drive through a chain barrier and rip a street sign out of the concrete. His candid assessment of the machines is that they often outperform the human drivers he sees every day. That is exactly what makes the enforcement gap so awkward.
Most traffic law assumes a person behind the wheel. Fault, citation, and correction all flow through a human being who can be identified, signed for, and held accountable. Remove the driver and the entire chain snaps. Texas is a clean example: police there cannot write a ticket for a car without a driver, because a citation requires a driver’s signature.
Why this is a story right now
The timing is not accidental. This week the head of the National Highway Traffic Safety Administration put the industry on notice. Administrator Jonathan Morrison sent AV developers a letter identifying what he called a clear pattern of driverless vehicles interfering with police, firefighters, and paramedics.
Morrison’s language was blunt. He described a car’s inability to recognize and respond to an emergency scene as a “functional insufficiency,” not a rare edge case, and gave companies until the end of July to bring him their fixes. The subtext is a regulator losing patience with a familiar industry defense: that failures around flares, cones, and flashing lights are unusual corner cases rather than a core competency the cars still lack.
The data: hundreds of incidents, a handful of consequences
Austin has become one of the country’s busiest robotaxi laboratories, and its records show the scale of the mismatch. Since July 2023, the city has logged 298 incidents involving autonomous vehicles, ranging from a car aimlessly circling a quiet neighborhood at night to a multi-vehicle wreck. Waymo vehicles accounted for 231 of them. Yet only eight of those cases resulted in officers filing a formal complaint.
That funnel is the whole story in one image. The incidents are common. The enforcement is close to nonexistent. Austin Police Lieutenant William White, who sits on several bodies focused on advancing autonomous technology safely, describes the complaint process as cumbersome enough that officers rarely bother relative to how easy writing a normal ticket would be.
The upside police actually acknowledge
It would be easy to frame this as machines running wild, but officers themselves push back on that. Autonomous vehicles usually drive less aggressively than people, cap their own top speeds, and never fail a breathalyzer. The companies point to internal data arguing their cars are statistically safer than human drivers. For a patrol officer, a fleet that does not speed, tailgate, or drive drunk removes a large share of the daily chaos.
The trouble is not recklessness. It is rigidity.
The real failure mode: the car that will not move
White’s biggest concern is not speeding robotaxis. It is frozen ones. He says the most common problem is a car that simply stops and refuses to adapt. In March, a Waymo in Austin blocked an ambulance trying to reach the scene of a shooting. The company said the delay did not affect patient outcomes and acknowledged it needs to move faster on the issue.
The pattern repeats nationally. Vehicles have stalled in front of fire stations, drifted into active construction zones despite visible cones, and failed to read the hand signals of an officer standing directly in front of them. On a highway near San Mateo, California, a Waymo carrying passengers reportedly kept rolling toward a construction zone while an officer shouted “Stop, Waymo” over a loudspeaker. A car that never speeds is small comfort when it cannot recognize that it needs to yield to a fire truck.
How cities are improvising
With the legal tools thin, police are getting creative. In Phoenix, officers went through Waymo’s in-person training to learn the slow, deliberate hand movements the cars can actually parse. Waymo says it has now trained more than 35,000 first responders nationwide. Phoenix has largely handled robotaxi missteps through direct communication with the company rather than paperwork, though a detective there notes that writing a citation is never fully off the table.
Other tactics are lower tech. At an Austin university graduation, several Waymo vehicles got stuck at a dead end. Oborski penned them in with traffic cones, built what he called a corral, and waited for Waymo to send humans to retrieve the cars. The most effective tool of all is prevention: near stadiums, concerts, and emergency scenes, police increasingly ask companies to geofence their cars out of the area entirely, a discipline that fixed, predictable routes make far easier to enforce.
A patchwork of state laws, changing fast
Because the United States has no centralized federal rulebook for how these cars should behave, enforcement is being invented state by state. California and Arizona already require robotaxi operators to have law enforcement interaction protocols before carrying passengers. Texas does not mandate that, though companies tend to offer police training anyway.
California just moved the goalposts. On July 1, a new rule built on Assembly Bill 1777 took effect, letting officers issue a “notice of autonomous vehicle noncompliance” directly to a manufacturer when a driverless car breaks the rules. Repeated or serious violations can lead the state to restrict or suspend a company’s permit. It is not a ticket in the traditional sense, but it is the first real lever that reaches past the missing driver and lands on the company. This is the same permission-by-patchwork dynamic that has stalled Waymo’s expansion into major markets like New York, where the wall is politics and regulation rather than technology.
What the companies are saying
Waymo, which remains the clear US leader with a fleet of roughly 4,000 vehicles, frames itself as wanting to be a public safety partner rather than a burden, and says its goal is to become a seamless part of city operations. It notes that its cars pull over for police lights and offer a way to reach a human representative. Zoox has publicly welcomed NHTSA’s push to hold the industry accountable. Tesla, whose upcoming Cybercab is a two-seat vehicle with no steering wheel or pedals built to respond to an officer’s hand signals, did not respond to requests for comment.
The gap between those tidy corporate statements and a Waymo blocking an ambulance is exactly what regulators are now trying to close.
What comes next
The cars are not slowing down, and neither is the buildout. Waymo already offers rides in 11 US cities and has announced expansion into Denver, Las Vegas, San Diego, and Tampa. Goldman Sachs projects nearly 63,000 commercial robotaxis on US roads by 2030. Every one of those vehicles is a rolling test of an enforcement system that was designed for human drivers and is being retrofitted on the fly.
The pressure points are clear: NHTSA wants solutions by the end of July, California’s new notices start generating a paper trail, and more states will follow with their own versions. The companies that treat first responder interaction as a first-class engineering problem rather than an edge case will have a real regulatory advantage. The ones that keep freezing in front of fire trucks will keep collecting notices, and eventually, suspended permits.
The Business Model Analyst Take
The enforcement headache is a symptom of a deeper business reality: in autonomous mobility, the hard, defensible value was never only the self-driving brain. As we have argued in our breakdown of the robotaxi cost stack, the intelligence layer is a small slice of the economics, while operations, fleet reliability, and yes, regulatory good standing, are where durable businesses get built or broken.
Right now the industry is optimizing for the wrong scoreboard. A car that drives 4,000 flawless miles and then blocks one ambulance does not generate a safety headline. It generates a trust deficit, a NHTSA letter, and a lever for regulators to slow your permit. For a company like Alphabet, which has told investors that Waymo will not be financially meaningful until 2027 or 2028, the near-term constraint on growth is not compute or capital. It is permission. And permission is granted by the same cities whose officers are currently penning robotaxis into cone corrals.
The winning move is counterintuitive but clear: treat first responders as the most important customer you are not charging. Master the emergency scene, the hand signal, and the two-minute geofence before mastering the next zip code. In a market with no national rulebook, the operator that makes itself the easiest for a police department to trust will scale fastest, because trust, not technology, is now the binding constraint.
