What Happens When Your Waymo Gets Attacked?

San Francisco resident Doug Fulop spent six minutes frozen in the backseat of a Waymo in January while a man punched at the windows and screamed at him and his two companions for “giving money to a robot.” Nobody took remote control of the car. Nobody drove up to intervene. Waymo Support told them to stay put and wait it out.

They did. Eventually the attacker got distracted, stepped back far enough that the car’s sensors registered a clear path, and the vehicle drove away. Police arrived later and confirmed Fulop’s account. He has not used Waymo at night since.

“As passengers, we deserve more safety than that if someone is trying to attack us,” Fulop said. “This can’t be the policy to be trapped there.”

It’s a fair complaint, and it arrives at an uncomfortable moment. Waymo is preparing to expand its robotaxi service to 20 cities this year, which means the company is about to introduce a lot of new riders to a product that, under certain circumstances, cannot protect them and cannot be overridden.

The core tension here is not hard to see. Waymo’s software is built around caution. The cars will not move if a person is too close. That’s a reasonable rule until the person who is too close is the threat. At that point, the feature becomes the problem. The car’s restraint traps the passengers as effectively as any lock.

A second incident, this one in Los Angeles, adds a different dimension to the story. Tech writer and speaker Anders Sorman-Nilsson was surrounded by five individuals on e-bikes who tried to open the doors of the Waymo he was riding in, banging on the windows and demanding he let them in. He refused. Sorman-Nilsson told reporters that, interestingly, he felt safer than he would have if a human driver had been behind the wheel, presumably because a human driver might have escalated or panicked.

That’s a useful data point, but it doesn’t resolve Fulop’s concern. The question isn’t whether a Waymo is safer than a human driver in most conditions. The question is what the car does when someone decides to use its own programming against it.

Anti-robotaxi sentiment in the Bay Area has been building for years, ranging from people blocking vehicles at crosswalks to take advantage of the sensor rules, to tagging and property damage, to the arson attacks on Waymo vehicles during anti-ICE protests in Los Angeles last June. As the fleet grows and the company enters new markets, these incidents are likely to become more frequent, not less.

Waymo has made its software somewhat more assertive over the years, and the cars drive more fluidly through traffic than they did when they first appeared on San Francisco streets. But the fundamental architecture still prioritizes not hitting anyone over extracting passengers from a dangerous situation. Whether that calculation is correct is a genuine ethical question, not a simple one.

Should the car have recognized an active threat and maneuvered away even at some risk of contact with the attacker? Should there be a remote-override option that lets a support agent take the wheel in an emergency? Should passengers have any physical controls available to them beyond a phone number to call?

Waymo has not publicly addressed any of these questions with specificity. The company has a support line. The support line told Fulop to stay in the vehicle. That’s currently the protocol.

The company’s planned expansion puts pressure on these unanswered questions. Riders in cities that have no experience with Waymo vehicles are going to encounter the same edge cases San Francisco has been generating for years, and they are going to expect answers that don’t yet exist.

Fulop’s situation lasted six minutes. That’s not a long time in most contexts. Trapped in a car while someone punches the windows and a crowd gathers outside, some of them cheering the attacker on, it’s a different matter entirely. The car eventually solved the problem by waiting. Passengers deserve a more deliberate answer than that.