Incident Overview
On January 18, 2026, a double-decker bus operated by the Spokane Transit Authority (STA) was directed by its onboard navigation system onto Cedar Street in downtown Spokane, Washington. The route led the bus under a low-clearance railroad viaduct, resulting in the shearing of the upper deck. The incident occurred during regular service and involved a total of 10 passengers, seven of whom were hospitalized with non-life-threatening injuries. The driver was later cited for negligent driving, though the primary cause of the incident was attributed to the automated navigation system’s routing error.
The navigation system in question was an onboard Computer-Aided Dispatch (CAD) mapping system provided by an unidentified vendor. This system is designed to optimize routes for efficiency but failed to account for physical infrastructure constraints, such as the height clearance of bridges and viaducts. The error highlights a critical limitation in automated routing systems: the absence of real-time or preloaded data on structural obstacles that could pose physical risks to vehicles and passengers.
Following the incident, the Spokane Transit Authority immediately suspended the use of CAD maps for routing purposes and temporarily pulled all double-decker buses from service. The agency launched an investigation to determine the root cause of the failure and to implement corrective measures. The suspension of the routing feature underscores the potential for automated systems to introduce safety-critical errors, even in the absence of a traditional machine learning failure or algorithmic bias.
The incident is notable for its direct physical harm to passengers, a consequence that stems from the interaction between automated decision-making and real-world infrastructure. Unlike many AI-related incidents that result in indirect harms—such as misinformation or privacy violations—this case demonstrates how routing errors in AI-driven systems can lead to immediate and tangible risks, including injury and property damage.
The Spokane Transit Authority’s response to the incident included internal warnings to operators to avoid relying on CAD maps for navigation. This directive reflects an acknowledgment that the system’s limitations were not immediately apparent to drivers or supervisors, despite its integration into daily operations. The citation issued to the driver for negligent driving may also indicate an attempt to address procedural failures, though the primary focus remains on the systemic issue posed by the navigation software.
Possible Root Cause
The root cause of the incident appears to lie in the design and implementation of the onboard CAD/navigation routing software. The system was developed by an unidentified vendor and deployed by the Spokane Transit Authority to optimize bus routes for efficiency. However, the software failed to incorporate critical physical constraints into its routing decisions, specifically the height clearance of bridges and viaducts along potential paths.
One plausible explanation for the error is the absence of comprehensive geographic data within the navigation system.
The incident does not appear to involve a traditional machine learning failure, such as a misclassified input or biased output. Instead, the error stems from a fundamental limitation in the system’s ability to recognize and avoid physical infrastructure risks.
The Spokane Transit Authority’s use of double-decker buses adds another layer to the root cause analysis. Double-decker buses are significantly taller than standard buses, increasing the likelihood of clearance issues in urban environments. The navigation system’s failure to account for this height differential suggests a lack of specialized configuration for vehicles of this type.
The vendor’s role in the incident remains unclear, as no specific details about the software’s development process, testing protocols, or post-deployment monitoring have been publicly disclosed. However, the incident serves as a case study in the potential consequences of assuming that automated systems will inherently account for all physical and operational constraints. The lack of transparency around the vendor’s involvement also raises questions about accountability and the need for clearer standards in AI system deployment, particularly in safety-critical applications.
Takeaways
The failure of the navigation system to account for physical infrastructure constraints resulted in direct harm to passengers, demonstrating that automated decision-making must prioritize safety over efficiency.
The case also highlights the importance of human oversight in AI-assisted operations. While automated systems can improve efficiency, they should not replace human judgment entirely, especially in environments where physical risks are present.
From a regulatory and industry standards perspective, the incident serves as a reminder of the gaps that exist in the oversight of AI systems in transportation.
The harm caused by this incident falls under multiple categories in the Domain Taxonomy of AI Risks, particularly within the domains of Human-Computer Interaction and AI System Safety, Failures & Limitations. The interaction between the navigation system and the driver resulted in a safety-critical error, while the system’s failure to account for physical constraints exemplifies its limitations in handling real-world complexities. This classification aligns with broader concerns about AI systems that operate in dynamic environments where human lives are at stake.
For organizations considering the adoption of AI-driven routing or navigation systems, this incident provides a cautionary example of the potential consequences of overlooking physical and operational constraints. Additionally, the involvement of double-decker buses in the incident points to the necessity of tailoring AI solutions to the specific requirements of the vehicles and environments in which they operate.
The Spokane Transit Authority’s proactive response—including the suspension of the problematic feature and the initiation of an investigation—sets a positive example for how to address AI-related failures transparently and responsibly.
Sources
Incident summary from the AI Incident Database, used under CC BY-SA 4.0.