Incident Overview
On February 27, 2026, the Washington State Department of Licensing (DOL) confirmed an incident where its AI-powered phone system failed to provide Spanish-language service to callers who explicitly requested it. Instead of routing callers to Spanish-speaking support, the system delivered AI-generated English responses spoken with a Spanish accent. This malfunction persisted for months before being corrected after public complaints and internal review.
The issue was reported by multiple callers, including Maya Edwards and her husband, who selected the Spanish self-service option (typically accessed by pressing '2') but received an AI voice speaking English in a heavy Spanish accent. The system’s failure to deliver actual Spanish-language assistance created confusion and frustration, particularly for non-English speakers relying on the service for licensing-related inquiries.
The DOL acknowledged the error and attributed it to a staff configuration mistake within the AI system. The agency temporarily disabled the Spanish-language feature while investigating the root cause. The incident highlights broader challenges in deploying AI-driven customer service tools, particularly in ensuring linguistic accessibility and avoiding unintended biases in automated interactions.
The AI system implicated in this failure was reportedly developed by Amazon and deployed by the DOL. While Polly is designed to support multiple languages, the misconfiguration in this case led to a breakdown in service delivery for Spanish-speaking callers.
The failure to provide accurate language support resulted in wrongful denial of services and potential discrimination against Spanish-speaking individuals, who were unable to access the intended assistance. The harm extended beyond individual callers to the broader public, undermining trust in government-provided AI services.
Possible Root Cause
The primary cause of the incident was a misconfiguration in the AI system’s language routing settings.
The use of Amazon Polly, a text-to-speech service, played a role in the malfunction. While Polly supports multiple languages and can simulate accents, the system was not configured to route callers to native Spanish-language content or human agents. Instead, it defaulted to an AI-generated English response delivered in a Spanish-accented voice, which failed to meet the callers’ expectations or needs.
The error persisted for an extended period, suggesting a lack of proactive monitoring or testing for language-specific configurations. Had the system undergone rigorous quality assurance checks—particularly for multilingual support—this flaw might have been detected earlier. The incident underscores the importance of thorough validation in AI deployments, especially in public-sector contexts where accessibility and equity are critical.
No evidence indicates malicious intent or deliberate exclusion of Spanish-language services. Instead, the failure appears to stem from an operational oversight, where technical implementation did not align with the intended user experience.
Takeaways
This incident serves as a case study in the risks of deploying AI-driven customer service systems without adequate linguistic and cultural considerations. The failure to provide accurate Spanish-language support demonstrates how misconfigurations in automated systems can lead to discrimination and exclusion, particularly for marginalized communities relying on public services.
The use of third-party AI tools like Amazon Polly does not absolve deploying agencies of responsibility for ensuring functionality and fairness. In this case, the DOL’s reliance on an AI system without proper oversight contributed to the prolonged disruption of Spanish-language services.
From a policy perspective, the incident reinforces the need for guidelines on AI accessibility in public-sector applications.
The harm caused by this incident—wrongful denial of services and potential discrimination—underscores the ethical implications of AI in service delivery.
Sources
Incident summary from the AI Incident Database, used under CC BY-SA 4.0.