Waymo Robotaxi Alerts Police to Firearm, Prompting Arrests and Privacy Questions

The arrests exposed unanswered questions about privacy and autonomous vehicles, including what the cars record, what happens to the footage and how much privacy riders should expect inside them. Waymo’s camera-covered cars have already been drawn into debates over privacy and surveillance. Bloomberg reported that law enforcement agencies have obtained search warrants for footage captured by Waymo cars on several occasions. Last year, the LAPD obtained footage of a hit-and-run recorded by a nearby Waymo vehicle. During anti-ICE protests in Los Angeles, five Waymos were vandalized and set on fire amid questions about whether the vehicles could capture images of protesters that might be used as evidence to identify or arrest them.

Waymo does not publicly disclose how many law enforcement requests it receives or how often it complies. Those concerns echo backlash against Flock’s network of automated license-plate readers, which have given police the ability to track cars across jurisdictions and have prompted growing fears about government surveillance and misuse. Unlike fixed Flock cameras, Waymos move through cities carrying cameras both outside and inside their vehicles.

The latest case takes the concerns a step further. Police did not approach Waymo after a crime looking for footage its cameras happened to capture. Waymo detected something inside the car and initiated the police response itself, according to the report.

Waymo spokesperson Julia Ilina told The Verge in an email that the company uses its internal cameras to help enforce its terms of service, including detecting prohibited items and behavior. She said Waymo uses automated technology to help its teams identify potential violations, such as the presence of firearms inside the vehicle. “If Waymo detects a firearm in the vehicle, we may reach out to local law enforcement,” she added.

Waymo did not explain how its automated system determines whether something may be a firearm, how confident it has to be before something gets flagged, what role humans play between an automated alert and a call to police, or what other violations of its terms of service might prompt a police response. This was not a human taxi driver deciding to call 911. It was a driverless car, equipped with cameras and an AI brain, escalating a situation involving its passengers to police.

Bob Somers, a former engineer at autonomous vehicle company Zoox and technical adviser to the AV-safety nonprofit SAVE-US, said he would expect Waymo’s system to look for a range of potentially dangerous weapons and behaviors. The software may not need to decide with certainty that it has seen a firearm, he said. It could instead flag something that resembles one for further review. “If there is a certain set of objects detected, then that would probably loop in a remote operator to take a look at what’s going on,” Somers said.

Phil Koopman, a Carnegie Mellon University professor emeritus who researches autonomous vehicle safety, said computer vision systems can be trained to recognize objects and potentially dangerous situations. “If you can train a computer to look at the road and decide how to drive, you can train a computer to decide if somebody’s holding a gun,” Koopman said. But those systems are “never 100 percent accurate,” he added.

That makes the unanswered question about human review important. A human taxi driver who sees a passenger holding something has context: they have been in the car for the entire ride, can hear what the passengers are saying, can ask questions and can decide whether the object is a real gun, a toy or something else. In July, Waymo reported two teens to police for underage drinking and shooting a toy gun loaded with water beads from the vehicle. It was not clear whether the teens ultimately faced any charges. A remote worker brought in to review a flagged moment might not have all of that context because they did not experience the ride firsthand, according to the report.