Autonomous vehicles, a dead-end street, and 53 minutes on hold — what happens when a private fleet of driverless cars becomes infrastructure-in-practice without being treated as infrastructure-in-planning.
These two incidents are separated by five months and very different in tone. One was a prank dreamed up by a tech hobbyist with a taste for systems satire. The other was a citywide emergency that exposed what happens when a private fleet of autonomous vehicles becomes infrastructure-in-practice without being treated as infrastructure-in-planning.
The prank matters because it demonstrated something important: a coordinated human action — no hacking, no laws broken, nothing more sophisticated than a ride-hailing app — could bring a portion of the Waymo network to a standstill and trigger a service suspension. The blackout matters because it showed what that same vulnerability looks like at scale, during an actual emergency, with real stakes.
For dispatch professionals, neither incident is primarily a technology story. Both are a communications and resource story. The question this exercise asks is the one your comm center will eventually have to answer: what do you do when a vehicle on your streets has no driver, belongs to a private company, is blocking an intersection or an emergency access route, and nobody at that company is picking up the phone?
What the comm center saw, and when. Color coding indicates the operational dimension.
Mary Ellen Carroll, Executive Director of San Francisco DEM, told the Board of Supervisors that first responders ended up becoming "default roadside assistance" for stalled Waymo vehicles — a role that pulled them away from life-safety calls. One firefighter spent time moving a stalled robotaxi out of an intersection instead of responding to a call. Supervisor Bilal Mahmood put it plainly: "What I'm hearing mostly is that you kind of still expect our first responders to do roadside assistance, and you are just going to help us train them better to do that."
In Austin, Texas, during a separate incident, a Waymo was filmed blocking a street that first responders were attempting to access during a mass shooting scene. The pattern across cities is consistent: autonomous vehicles, when they fail, fail in the roadway, often at intersections, and they require some form of human intervention to move — intervention that either comes from a remote Waymo operator (reachable by phone, eventually) or from a first responder on scene who has now become an unplanned tow truck driver.
After Waymo phased out in-car safety monitors in June 2024, the San Francisco Municipal Transportation Agency was forced to create a new category for emergency and transit dispatchers: "Driverless Car Incidents." Average call time from the Traffic Management Center to Waymo to resolve a stall: approximately 20 minutes. Your center may not have Waymos yet. But the category is coming.
Vendor hotlines are designed for isolated incidents, not system-wide failures. The same event that requires you to call the hotline is likely the same event overwhelming the hotline. The hotline is your last resort, not your plan. The plan is what you do when nobody picks up.
The legal-but-deliberate disruption is a real attack surface. 50 people with smartphones briefly disabled Waymo service in a 2-block area using nothing more sophisticated than a ride-hailing app. The DDoS prank is the proof of concept; coordinated manipulation of autonomous systems during a crisis is a planning question for emergency management.
First responders are not the AV operator's roadside assistance. Every minute a firefighter spends moving a stalled robotaxi is a minute not available for a medical call, fire, or extrication. The framing question — whose responsibility — is an advocacy conversation, not just operational.
Build the protocol before deployment, not after. A CAD code for AV incidents, a verified first-responder contact above the public hotline, an escalation protocol when that contact is unreachable, and a conversation with your EM office about geofencing or service-suspension capabilities. Most centers will not build this until after their first incident — but the San Francisco experience gives you a roadmap.
Training requests work both directions. Waymo has trained 25,000+ first responders nationally on vehicle interaction. If your jurisdiction is in an AV expansion market, request that training proactively — not because managing their equipment is your responsibility, but because your responders will encounter the vehicles regardless of who is technically responsible.
No right answers. Tap a question to expand the analysis. Use one or all — whatever fits your time.
Most centers do not yet have a specific call type for this scenario. It likely gets logged as a traffic hazard, abandoned vehicle, or obstruction — none of which capture what it actually is.
That matters because the response is different. You cannot dispatch a tow truck to an autonomous vehicle without the owner company's involvement. You cannot tell the driver to move it. Your traffic unit cannot write a citation to the car.
San Francisco had to create a new CAD category called "Driverless Car Incidents" specifically because existing categories did not fit the problem.
The question for your agency: do you have autonomous vehicles operating in your jurisdiction now? If yes, what is your call type, and does your dispatcher know what number to call? If no, when does that change, and when does your protocol get updated? The phone number for Waymo's first responder line, for reference: 1-877-503-0840. That is also the number that put a San Francisco dispatcher on hold for 53 minutes during a blackout.
Vendor hotlines are designed for isolated incidents. They are not scaled for simultaneous system-wide failures. The same event that requires you to call the hotline is likely the same event that is overwhelming the hotline.
This is structurally true for autonomous vehicle companies, for CAD system vendors, for utility companies, and for any private system that embeds itself in public infrastructure without a public-sector redundancy.
The lesson is: the hotline is your last resort, not your plan. The plan is what you do when nobody picks up. Does your center have a protocol for when a critical vendor is unreachable during a major incident?
San Francisco's solution was to have the mayor personally text the CEO. That is not a protocol — that is improvisation at the executive level. The gap between what the law requires (30 seconds for emergency geofencing response under a proposed California bill) and what actually happened (53 minutes on hold) is where your planning lives.
The Walz prank was a novelty. But it demonstrated a real attack surface: if 50 people with smartphones can briefly disable Waymo service in a 2-block area, a larger coordinated action with worse intentions could do significantly more.
The prank was the proof of concept. The question for emergency management is whether similar manipulation of autonomous systems — ride-share, delivery, logistics — could be used deliberately during a crisis to slow emergency response or create diversionary incidents.
This is not a hypothetical that needs a policy answer today. But it is a conversation worth having with your jurisdiction's emergency management office before autonomous vehicle density in your city makes it operationally relevant.
The communication between private autonomous vehicle operators and public safety agencies is still largely informal in most cities. The DDoS prank is a useful training hook because it is nonthreatening enough to discuss openly, while pointing directly at a real vulnerability.
Supervisor Mahmood nailed it at the hearing: Waymo's answer was essentially "we will train your first responders to do our roadside assistance better." That is not a satisfactory answer.
First responders are not Waymo employees. Every minute a firefighter spends moving a stalled robotaxi is a minute they are not available for a medical call, a fire, or an extrication.
For dispatch professionals, this is an advocacy conversation as much as an operational one. If autonomous vehicles are going to operate in your jurisdiction, the question of who bears the cost of their failures needs to be answered before the failure happens. Does your agency have a formal relationship with any autonomous vehicle operators in your jurisdiction? Is there an MOU, a direct contact above the hotline level, an agreed-upon protocol for removal during an active incident? If the answer is no, the next blackout will play out the same way.
The training point for dispatchers: if you receive a call about a stalled autonomous vehicle blocking emergency access, escalate to your supervisor immediately rather than attempting to resolve it through the standard hotline process. Time is a factor, and the standard process may not produce a result in time to matter.
The honest answer is that most centers will not prepare until after the first incident. But the San Francisco experience gives you a roadmap for what to build before you need it.
At minimum: a call type or CAD code for autonomous vehicle incidents; a verified first responder contact above the public hotline level for any AV operator in your jurisdiction; a protocol for escalation when that contact is unreachable; and a conversation with your jurisdiction's emergency management office about what geofencing or service suspension capabilities exist and how they are activated.
The harder conversation is the one about training. Waymo has trained 25,000+ first responders across the US on how to interact with its vehicles. If your jurisdiction is in an expansion market, you should be requesting that training proactively — not because it is your responsibility to learn how to manage their equipment, but because your responders will encounter these vehicles regardless of who is technically responsible.
The San Francisco model for what not to do: wait until a blackout creates 1,593 simultaneous stall events, then have the mayor text the CEO.
Five questions. Answer, then submit for inline feedback. Progress saves locally.