Urban Autonomy Lab
Chapter 4 · EV110

Waymo Suspends All Freeway Rides in Four Cities After Robotaxi Speeds Through a Closed-Freeway Construction Zone — 21 May 2026

Date collected: 25 May 2026
Summary

All freeway rides suspended in four cities after a robotaxi sped through a closed-freeway work zone and was chased by police; dynamic ODD features unsolved at high speed

Headline Finding

On 21 May 2026 Waymo suspended all freeway rides across San Francisco, Los Angeles, Phoenix, and Miami while it works to improve its vehicles’ performance in construction zones, the company confirmed to TechCrunch. Surface-street service continued. The suspension followed a 19 May rider account, posted with video, describing a Waymo that “freaked out and sped up to highway speeds through construction trucks” on a closed section of freeway, after which the vehicle was reportedly chased by police; the rider wrote, “Genuinely thought we were about to die.” This is a distinct ODD failure surface from the flood sequence (EV109): temporary, dynamic roadway features — work zones, lane closures, temporary signage — encountered at the highest-speed, highest-consequence operating mode. It is the second correlated capability withdrawal by a single operator across multiple cities in the same week, and it concerns the operating mode on which Waymo’s airport-connection and trip-time value proposition most depends.

Detail

  • Source: TechCrunch (Sean O’Kane) 21 May 2026; The Next Web 23 May 2026; rider posts on X dated 19 May 2026. Window: 19–25 May 2026.
  • Action: freeway/highway ride service suspended in San Francisco, Los Angeles, Phoenix, and Miami. Surface-street robotaxi service continued in those cities.
  • Trigger: Waymo did not cite a specific incident but confirmed it is evaluating freeway performance in construction zones. A rider (@Elliot_slade) posted on 19 May that the robotaxi “blasted through cones” and “sped up to highway speeds through construction trucks” on a closed freeway segment, claiming the vehicle was chased by police.
  • Company statement: Waymo said it is “integrating recent technical learnings into our software and expect to resume these routes soon,” with no timeline given.
  • Context on the freeway capability: Waymo began offering freeway rides in late 2025 in three cities. Freeway access is central to its expansion in large metros — it connects riders to airports and cuts trip times (in the Bay Area, cross-peninsula trips that previously took 45 minutes to over an hour). Waymo is concurrently testing its Zeekr-built sixth-generation vehicle (“Ojai”), with rides expected in the coming months.
  • Concurrency: the freeway suspension ran in parallel with the five-city flood-related pause (EV109), so that in the same week a single operator both withdrew an entire functional capability (all freeway driving in four cities) and suspended whole-city service in five others.

Governance Implication

Four points for the paper.

  1. Dynamic ODD features are an unsolved high-speed failure surface, and the Institutional ODD framework (Ch.3) should treat them as a distinct authorisation gate. Static map features can be pre-encoded; construction zones change daily, are not in HD maps, and are exactly where a mapping-and-rules architecture is weakest (the same architectural limitation The Next Web identifies in the flood case). The paper should argue that cities and state DOTs must require demonstrated competence in temporary/dynamic ODD features — work zones, lane closures, temporary traffic control — as a precondition for high-speed (freeway) authorisation, separately from baseline surface-street authorisation.

  2. This is a data-sharing governance hook (Ch.6 Sensing). Work-zone permits are issued by city and state transport authorities, which therefore hold authoritative, near-real-time data on where active construction zones are. Operators demonstrably lack reliable access to it. A municipal/state requirement to publish active work-zone and lane-closure feeds — and a reciprocal requirement that operators consume them — is a concrete, low-cost instrument that directly addresses the failure mode.

  3. The police chase of an autonomous vehicle on a closed freeway is a first-responder-interaction failure. It connects to the Texas SB 2807 first-responder-plan requirement (EV107) and the Austin first-responder incidents (EV92), and underlines that first-responder protocols designed for low-speed surface-street stalls do not cover a vehicle behaving erratically at speed on a restricted carriageway. Ch.5 (Allocation) and Ch.7 should treat high-speed first-responder interaction as a separate protocol requirement.

  4. Commercial and legitimacy signal. Freeway driving is the capability most load-bearing for Waymo’s stated scaling narrative (one million rides per week, World Cup airport connections). Suspending it across four cities — the highest-consequence operating mode proving the least mature — is direct evidence that the deployment-pace problem the paper centres on (Ch.1) extends into the operating envelope itself: capability is being marketed and scaled ahead of where it is reliable. Ch.5’s Legitimacy treatment should pair this with EV109 as a same-week demonstration that correlated, operator-initiated capability withdrawal is now a routine feature of the market.

Sources

Thread