Waymo Tap-Dances about Overseas Remote Drivers

They have remote human drivers in a very real sense, but they aren't full teleoperators

Feb 05, 2026

Since I have received several press queries on this question, this week I’ll do a slightly early posting with an update on Waymo’s remote assistants. Waymo has in the past characterized their human remote assistants as a sort of phone-of-friend arrangement. And now we have confirmation that at least some of those friends are overseas. But are they really drivers?

When called out on this specifically in a US Senate committee hearing yesterday, Waymo tap-danced as to whether that person is a driver, how many such people are overseas, and so on. I hope journalists dive deeper here, because if they are actually drivers there are important questions about qualification, licensure, jurisdiction, and accountability that need to be answered.

For today I will make some observations as someone who has watched lawyers at work on highly technical cases and seen a lot of subtle deflection to important questions:

🔹 Waymo said “the Waymo vehicle is always in charge of the dynamic driving task.” That is not a denial that the human is performing some of the driving task. (Being “in charge” is not the same as actually performing.)

🔹 Waymo said about those remote humans “they provide guidance; they do not remotely drive the vehicles.” If pressed, that can be cast as a denial of full teleoperation, but does not get to whether the remote assistant is completing the dynamic driving task by, for example, providing the color of a traffic light to the vehicle, and therefore fulfilling a driving role. (Hint: they are, and that specific example was involved in a reported Waymo mishap described in my below-referenced Substack piece.)

🔹 Waymo said of the remote assistant information, “that is just one additional input.” So is an accelerator pedal, which for decades now has been just one computer input among many in determining delivered wheel torque. (I’m not saying their remote drivers have an accelerator. I’m pointing out that providing “one additional input” is sufficient to make someone a driver.)

Waymo did not actually say their remote assistants in the Philippines are drivers. But they did not actually deny it either when they had a golden opportunity to do so. They dodged the question. If Waymo has actually said in writing “our remote assistants are not drivers” somewhere (anywhere) that would have a legally binding effect on the company, let me know. I’d love to see it. But I didn’t hear it happen in this hearing.

A case could (perhaps should) be made that Waymo vehicles revert to SAE Level 2 for a few tens of seconds when asking for remote assistance, then resume Level 4 operation. SAE J3016 clearly permits mode shifts of that type mid-journey. And Waymo has yet to deny this. Perhaps because they cannot truthfully do that.

The truth is that All Robotaxis Have Remote Drivers, and the AV industry wants to kick the accountability can for that down the road as long as they can. I suppose they keep thinking those remote assistants will evaporate any month now. But they were wrong about it a year ago when I wrote an article on the topic, and they are wrong about it now.

The question is not whether they will have remote driver assistants, but rather whether the safety of the robotaxi+remote assistant combination is acceptable. Which was what the Senator was driving at.

Article here: All Robotaxis Have Remote Drivers (link)

The video segment of the hearing is here. It is worth listening to the full exchange with Senator Markey at time 1:28:52 (direct link to that time stamp)

Full hearing below:

Update 2/5/2026:

A friend pointed out this statement from Waymo in their help pages: "Waymo doesn’t operate any of its cars remotely — when in autonomous mode, the car is responsible for its own driving at all times." Source: https://support.google.com/waymo/answer/9699657

Superficially this seems like a stronger statement, but let's break this one down too. To be clear, Waymo is not claiming this authoritative, but it is another good example of the rhetorical devices they employ to dodge this issue.

"doesn’t operate any of its cars remotely" => Not full teleoperation. Stipulated, but not on point to the question at hand.

"when in autonomous mode" => When it phones a friend, it is not really in autonomous mode, is it? So not relevant to the question at hand.

"car is responsible for its own driving at all times" => A car cannot be legally responsible for anything, because it is not a legal person. So this is nonsense.

Phil Koopman has been working on self-driving car safety for about 30 years, and embedded systems for even longer. For more on applying AI, see his new book: Embodied AI Safety.