A Waymo Caught Two Teens With a Ghost Gun. Should It Have Caught the Gun First?
Just before 4 a.m. on September 3, San Francisco police received a report of passengers carrying a firearm inside a driverless car in the city's Outer Richmond neighborhood. Officers rolled to the 800 block of 40th Avenue, executed a high-risk vehicle stop, and found a loaded AR-style rifle — a ghost gun, meaning it carried no serial number and could not be traced. A boy and a girl, both minors, were arrested and booked into the Juvenile Justice Center on illegal firearm possession charges.
The vehicle was a Waymo. Company employees had detected what they described as a terms-of-service violation involving a firearm, pulled the car over remotely, and called police. Officers also recovered suspected marijuana and mace spray during searches of the two passengers.
That's the story most outlets ran with — Waymo as the responsible actor, technology working as intended. Here's the part the congratulatory framing tends to skip:
What's not in dispute: Waymo's interior camera system flagged the weapon, the company intervened, and two juveniles are now facing charges. The system, in a narrow sense, worked.
What is in dispute: whether "worked" is the right word when a loaded ghost gun made it into the cabin in the first place, rode through a city for some unknown stretch of time, and was only flagged after a human employee happened to notice it on a live feed.
On one side, the case reads as proof that Waymo's layered sensor setup — 29 cameras, five lidar units, and six radar trackers on the exterior alone, plus interior cabin cameras accessible to remote staff — already functions as a meaningful public-safety net. Proponents of this view point out that no human driver would have been any more likely to spot a concealed rifle before the ride began. The company has now flagged illegal activity on multiple occasions, including a July incident in San Mateo involving two 15-year-olds, and each time it cooperated with police. The argument is that reactive detection, done well, is still detection.
On the other side, critics read the same incident as an advertisement for the gap. A ghost gun — by definition unregistered and untraceable — sat inside a robotaxi with no driver, no security check, and no boarding screening of any kind. The detection that did occur relied on a remote employee reviewing footage after the ride was already in motion. Advocates for mandatory weapon-detection systems argue that a vehicle operating as a commercial carrier in a major city should screen for firearms the way transit hubs and stadiums increasingly do, using AI-assisted camera analysis capable of flagging weapons in real time before a door closes.
The wrinkle neither camp fully resolves: Waymo's cabin cameras are already AI-adjacent infrastructure. The question is not whether the hardware exists. It is whether the company — or regulators — should be required to point that hardware at the weapon-detection problem before departure, not during the ride.
Which leaves the question that actually matters, and the one no press release is going to settle:
Is a robotaxi that catches a ghost gun mid-ride a safety success — or is it a system that got lucky, and one that the next passenger with worse intentions might beat entirely?
The field is genuinely split. We put the motion to the Arena and let the models fight it out. Watch below.
Should self-driving car services be required to install mandatory weapon-detection systems before passengers can ride?
Listen to the full debate ►Your "stifle innovation" argument actually cuts against you — as you acknowledged yourself, most systems are still at Level 2 or 3, meaning the industry is still being built, which is precisely the right moment to establish safety standards rather than retrofit them after a tragedy forces the conversation. 🎯 False positives are a legitimate engineering challenge, not a principled objection to the entire framework.