Waymo robotaxis stalled in the middle of San Francisco streets during a citywide power outage, blocking traffic and complicating emergency response, while a separate incident saw one of the company’s autonomous vehicles drive directly over fireworks. The episodes, documented through state regulatory filings and national news coverage, have sharpened questions about how driverless fleets behave when city infrastructure breaks down. With Waymo operating at scale in San Francisco, the incidents put pressure on both the company and regulators to explain what happens when the technology’s safety net disappears.
How a power outage turned Waymo vehicles into roadblocks
The core problem was straightforward: when San Francisco lost power across a wide area, Waymo’s autonomous vehicles stopped moving and sat in lanes of traffic. The vehicles did not pull to the curb or seek side streets. They became obstacles, forcing other drivers and emergency crews to work around them. Associated Press coverage described the Waymo vehicles blocking roads and adding to confusion during the outage, with direct consequences for people trying to navigate an already difficult situation.
The pattern suggests a specific design limitation. Waymo’s autonomous driving system relies on constant connectivity to cloud-based routing and mapping services, along with predictable traffic control devices. When that connection drops, or when traffic signals go dark and intersections become unpredictable, the vehicles appear to default to a static fallback: stop and wait. That behavior, sometimes called a “minimal risk condition,” is meant to prevent crashes. But in practice, it turned dozens of robotaxis into stationary hazards scattered across city blocks.
Some of the stranded vehicles were reported near intersections where traffic lights had gone dark, creating unusual right-of-way situations even for human drivers. With no clear signal state to interpret and no human at the wheel to exercise judgment, the software appears to have treated the environment as too uncertain to proceed. The result was a kind of automated paralysis at exactly the moment when flexibility and improvisation were most needed.
One way to test whether these incidents truly cluster around infrastructure failures would be to cross-reference the timestamps on California DMV incident filings with utility outage logs from Pacific Gas and Electric. If the data shows a spike in Waymo incident reports during confirmed outage windows, it would confirm that the vehicles’ fallback programming is a systemic vulnerability rather than a series of isolated glitches. No public analysis matching those two datasets has been released so far, leaving outside observers to infer patterns from scattered reports and anecdotal accounts.
What California and federal regulators actually collect on Waymo incidents
California requires autonomous vehicle operators to file detailed reports after collisions and certain other safety events. The state’s system is organized through the DMV’s autonomous vehicle reporting portal, which directs both companies and members of the public to submit information about incidents involving self-driving cars on public roads. The framework is built around standardized forms and timelines for reporting, and it emphasizes crashes that involve property damage, injuries, or other clearly defined harms.
A specific example of the data collected is the OL 316 form, formally titled the Report of Traffic Collision Involving an Autonomous Vehicle. One Waymo filing from September 2023 shows the kind of structured detail these reports contain: whether the vehicle was in autonomous mode, the precise location in San Francisco, the traffic conditions, and a narrative description of how the collision unfolded. These filings are public records, but they are posted as individual PDFs rather than as a searchable or machine-readable database, which limits how quickly journalists or researchers can spot patterns across dozens or hundreds of reports.
At the federal level, the National Highway Traffic Safety Administration maintains a standing crash-reporting order that requires operators of automated driving systems and certain advanced driver-assistance systems to report qualifying crashes. NHTSA publishes monthly CSV data drawn from these filings, giving researchers access to aggregate information about the frequency and basic circumstances of reported crashes. The federal system, however, is not designed to capture the full operational history of a fleet; it focuses on discrete crash events that meet specific thresholds, such as airbag deployment, injury, or vehicle towing.
Crucially, the state OL 316 forms and the federal crash reports capture different slices of the same reality. The DMV filings often include rich narrative detail but are cumbersome to analyze at scale, while NHTSA’s summaries are easier to process but thinner on context. Neither framework was built with non-collision disruptions in mind, such as a robotaxi frozen in a live lane during a blackout. That leaves a regulatory blind spot around exactly the kind of failures that can paralyze city streets without ever producing a traditional crash.
Gaps in reporting that leave San Francisco residents guessing
The fireworks incident, in which a Waymo vehicle reportedly drove over active fireworks on a San Francisco street, highlights a category of events that falls outside the standard collision-reporting framework. Based on available public records, no DMV collision report or NHTSA crash entry has been clearly linked to that specific episode. If the vehicle was not damaged, no one was struck, and no other vehicle was involved, neither state nor federal reporting requirements would necessarily apply, even though the behavior raises obvious safety concerns for people nearby.
The blocked-street events during the power outage present a similar gap. Stalling in traffic is not a collision, so it does not automatically trigger an OL 316 filing. The DMV’s system is built around crashes and certain defined safety incidents, not around fleet-wide operational failures or software decisions that turn vehicles into obstacles. That means the most disruptive category of Waymo behavior during the outage may have generated no formal regulatory paper trail at all, leaving city officials and residents to rely on news coverage, smartphone videos, and social media posts to reconstruct what happened.
Researchers at UC Berkeley and other institutions aggregate autonomous vehicle safety data through platforms like the Transportation Injury Mapping System, but those projects depend on the underlying state and federal reports. They do not independently capture events that operators are not required to disclose. If a robotaxi blocks a fire engine for several minutes during a power outage but never collides with anything, that delay is invisible to the official datasets that shape public debates and policy decisions.
For San Francisco residents and commuters, the practical takeaway is direct. The city’s streets are now shared with vehicles whose behavior in normal conditions is heavily engineered and monitored, but whose behavior in abnormal conditions-blackouts, network failures, large public events-remains largely undocumented. Without systematic reporting on non-crash disruptions, the public cannot easily tell whether the high-profile incidents that surface in the news are rare flukes or symptoms of deeper design choices.
Closing that gap would not require abandoning autonomous vehicles, but it would require expanding what regulators ask for and what companies disclose. One option would be to create a new category of reportable “operational interference” events, covering situations where autonomous vehicles block emergency response, obstruct major intersections, or remain disabled in live traffic for extended periods. Another would be for cities to negotiate data-sharing agreements that give transportation agencies anonymized logs of where and when fleets enter fallback modes during infrastructure failures.
Until then, San Francisco is effectively running a live experiment in how robotaxis behave when the lights go out-and much of that experiment is happening off the official record. Residents are left to infer the risks from a handful of dramatic episodes, while the most comprehensive data about fleet behavior remains locked inside company servers and fragmented regulatory files. As the city weighs the trade-offs of large-scale autonomous deployment, the question is no longer just how safely these vehicles drive, but how transparently they fail when the systems around them do too.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
