Thirteen Waymo Robotaxis Drove Into Live Construction Zones Before Anyone Could Stop Them
What happened
Between April 11 and May 18, 2026, thirteen Waymo robotaxis entered closed or active freeway construction zones in Phoenix and the San Francisco Bay Area. The vehicles did not stop at the closure controls. They continued into active work areas at freeway speeds, putting construction workers in the path of vehicles moving at highway velocity with no human in the seat to override what the automated system had already decided.
The root cause, according to reports, was a failure to recognize construction-zone closure controls. These are not subtle cues. Closed work zones on freeways are marked with barriers, cones, signage, lane-redirect arrows, and often human flag workers standing in the roadway. Any of those signals should communicate to an approaching vehicle that the road ahead is not open. Instead, the automated driving systems on these thirteen vehicles read what they encountered as navigable road and proceeded.
Work zones are a known hard case for automated driving. The geometry is non-standard, the signage is temporary, barriers can be placed inconsistently across job sites, and flag workers occupy positions that do not fit the lane-and-lane-marking model a system optimized for normal freeway conditions is built around. A vehicle can encounter a closed construction zone and parse it incorrectly, not because something broke but because the system was never reliably taught what a closed work zone looks like across the full range of configurations crews actually use. The result at freeway speed is not a slow-moving collision risk. It is a vehicle doing sixty or seventy miles per hour through a space where the ordinary rules about where cars belong have been suspended.
Waymo's response moved in two stages. The company restricted freeway operations while it investigated, then issued a recall covering 3,871 automated driving systems. A recall at that scale is not a single-vehicle anomaly. It confirms a systemic deficiency that required a fleet-level fix.
The incidents accumulated over five weeks before that recall was announced, thirteen separate incursions across two states with construction crews present each time. The question the timeline raises is whether the pattern was visible internally before it triggered a public response. A provable record of what each system perceived at the moment of each incursion, what decision it logged, and when those logs surfaced for human review would make it possible to determine whether the five-week window reflected the speed of discovery or something slower. Without that record, accountability for automated driving failures begins at the recall announcement rather than at the first vehicle that passed through a closure it should have read as closed.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.