Urban Autonomy Lab
Chapter 2 · EV89

China MIIT Suspends Issuance of New Robotaxi Permits Nationwide Following Baidu Apollo Go Wuhan Cloud-Outage Fleet Stop — 29 April 2026

Date collected: 4 May 2026
Summary

First nationally-imposed pause on robotaxi permitting globally; fallback-capability requirement now formally regulated

Headline Finding

On 29 April 2026, Bloomberg, CnEVPost, and Nikkei Asia reported that China’s central authorities have suspended the issuance of new autonomous-vehicle permits nationwide following the 1 April Baidu Apollo Go cloud-service outage in Wuhan, in which more than 100 robotaxis stopped mid-route after losing communication with their centralised dispatch platform. Operators and local governments are now required to demonstrate improved safety measures — specifically, the ability for vehicles to complete a trip or return to a parking area independently if the cloud connection drops — before new licences will be issued. A final regulatory decision is not expected before the end of May. This is the first nationally-imposed regulatory pause on robotaxi permitting that has been triggered by a fleet-coherence event rather than by a single-vehicle incident.

Detail

  • Triggering event (1 April 2026): A Baidu cloud anomaly severed communication between the Apollo Go fleet in Wuhan and the centralised dispatch platform. With no on-board fallback for completing a trip or safely parking, more than 100 vehicles stopped in lane simultaneously, trapping passengers and disrupting traffic. (See EV49 for the original incident narrative; the 29 April action is the regulatory follow-on.)
  • Regulatory action (29 April 2026): Reported by Bloomberg (“China Suspends Autonomous Driving Permits After Baidu Outage”), Nikkei Asia, CnEVPost, and electrive. Authorities have halted new robotaxi licence issuance pending fleet-level safety improvements. Existing operations continue, but operators cannot bring additional vehicles into service, kick off new test programmes, or expand into new cities while the suspension remains in force.
  • Specific technical fix demanded: Vehicles must be able to complete a trip or return to a parking area independently if the cloud-connection drops. This is an explicit regulatory translation of the failure mode into a fallback-capability requirement.
  • Operational scope: The suspension applies nationally and to all operators (Baidu Apollo Go, Pony.ai, WeRide, AutoX, Momenta, CaoCao). Apollo Go’s Wuhan operations specifically remain suspended under local-authority investigation.
  • Timing: A final decision on permitting status is not expected before the end of May, per Nikkei Asia. Carscoops and others framed the freeze as the “first nationwide robotaxi licensing pause anywhere globally.”
  • Capital-market signal: Stocktwits and other outlets report Baidu (BIDU) shares under pressure on the news.

Governance Implication

This is the cleanest possible empirical confirmation of the Fleet Coherence sub-mode of the Allocation failure mode named in the paper. Three implications:

  1. Fleet Coherence has now been operationalised in regulation. The MIIT directive translates the abstract failure mode (simultaneous fleet-level halt due to centralised-control dependency) into a specific permitting precondition (on-board trip-completion or self-park fallback). The paper’s Ch.5 should treat the 29 April action as the first regulatory artefact directly addressing fleet coherence as a distinct failure category. No NHTSA, NTSB, UNECE, or DfT instrument has yet done so.
  2. Cross-reference for the Governability Stack — Data × Rules cell. The fix targets the data-architecture dependency between vehicle and cloud as the regulated object. This validates the Stack’s separation of Data from Sensing — the failure was not a sensor or perception failure, it was a data-availability failure. The paper’s Ch.6 (Sensing) and Ch.9 (Regulatory Horizon) treatment of fleet-level data-architecture risk now has a named, dated, jurisdictionally-binding worked example.
  3. Pacing-Problem signal in the opposite direction from the US. The 29 April action is a deceleration of deployment in the largest robotaxi-vehicle market by fleet size. Combined with EV82 (Mobileye Q1 Chinese-OEM upside) and EV79 (CaoCao Hangzhou + 100k-target), this introduces a fourth pacing dimension: regulatory pace of fleet-level safety responses. China is the first jurisdiction to demonstrate that this pace can be fast, central, and binding when triggered by a coherence event. The paper’s Ch.1 Pacing Problem section now has a non-Western precedent for proactive city-or-state-level governance — the policy mechanism the paper argues for.

The paper should also flag that, by mid-May–end-May, the international comparison case will be sharpened: how Chinese regulators re-open permitting (with what conditions) is the second-order data point. If the precondition for re-opening is the fallback-capability requirement, Western jurisdictions will face a clear question of whether their permitting frameworks demand equivalent on-board fallback or rely instead on operator self-attestation of safety cases.

Sources

Thread