Seattle Fire Department Secretly Deploys AI on 911 Calls

When citizens dial emergency services during a medical crisis, they expect human judgment to guide their response. Instead, a growing number of municipalities are quietly integrating artificial intelligence into the dispatch workflow. In Seattle, the fire department has utilized a real-time monitoring system to analyze every medical 911 call for over a year without notifying the public or securing formal regulatory approval. This undisclosed deployment raises fundamental questions about transparency, public trust, and the boundaries of municipal surveillance.

Seattle Fire Department has secretly deployed Corti AI to monitor all 911 medical calls since December 2023, prompting dispatchers to route callers to a nurse line without public disclosure or surveillance ordinance review. This undisclosed implementation highlights significant gaps in municipal oversight and raises urgent questions about algorithmic transparency in public safety.

What is the Corti AI system and how does it operate within emergency dispatch?

The Copenhagen-based startup Corti developed a sophisticated artificial intelligence platform designed to assist emergency medical dispatchers. The company secured sixty million dollars during a Series B funding round to expand its operations across multiple international markets. Seattle Fire Department initially partnered with the vendor in twenty twenty-one to explore triage support capabilities. The specific implementation of live artificial intelligence prompting to redirect callers began in December of twenty twenty three. Dispatchers receive real-time suggestions to route certain individuals away from standard emergency response protocols.

The system continuously analyzes incoming medical calls to identify patterns that might indicate lower-acuity conditions. When the algorithm detects specific linguistic markers, it generates a prompt for the dispatcher. These prompts suggest diverting the caller to a nurse consultation line operated out of Texas. The fire department maintains that dispatchers retain final authority over every routing decision. However, the presence of automated suggestions fundamentally alters the traditional call-handling workflow. The technology does not store call audio recordings, and it does not attempt to identify individual callers by name.

This operational model differs significantly from traditional dispatch practices. Historically, human operators relied on standardized medical protocols to assess caller severity. The new approach introduces a layer of algorithmic judgment that functions as an invisible intermediary between the citizen and the emergency responder. While the department argues that the system merely assists human decision-making, the continuous real-time analysis of private conversations establishes a new precedent for municipal technology deployment. The lack of public notification means callers have no awareness that an algorithm is evaluating their medical distress.

The underlying architecture relies on natural language processing techniques to parse spoken medical complaints. The algorithm scans for specific keywords, vocal stress indicators, and contextual patterns associated with non-urgent conditions. When the system reaches a confidence threshold, it generates a visual prompt for the dispatcher. This automated triage aims to reduce unnecessary ambulance deployments and alleviate strain on emergency medical resources. The technology represents a significant departure from traditional human-led assessment protocols.

The operational framework requires continuous monitoring of audio streams without retaining the recordings. This design choice was explicitly cited by the fire department as justification for bypassing surveillance regulations. The system processes data in real time and immediately discards the audio input after generating routing suggestions. While this approach minimizes long-term data storage risks, it does not eliminate concerns about continuous monitoring and algorithmic influence. The distinction between temporary processing and permanent recording remains legally contested in municipal governance circles.

Why does the lack of public disclosure matter for civil liberties?

Seattle established a comprehensive surveillance ordinance in twenty seventeen to address growing concerns about municipal data collection. The legislation, codified as section fourteen point eighteen of the municipal code, requires city departments to obtain explicit city council approval before deploying technology that observes or monitors individuals. The ordinance was designed to provide the public with a direct voice regarding how government agencies utilize monitoring tools. The fire department deliberately bypassed this review process by classifying the Corti system as a decision-support tool rather than a surveillance technology.

Legal experts argue that this classification ignores the practical impact of the technology on individual rights. Ryan Calo, a law professor at the University of Washington, emphasizes that citizens routed away from emergency response deserve to understand the mechanisms behind that decision. The distinction between surveillance and decision-support remains legally ambiguous within the current ordinance framework. Courts have not yet tested whether real-time analysis of emergency communications qualifies as surveillance under existing municipal law. This ambiguity allows departments to deploy powerful monitoring systems without triggering statutory oversight requirements.

The transparency gap creates a significant governance problem that extends beyond technical implementation. When public institutions deploy artificial intelligence in healthcare settings, the information asymmetry between the agency and the public undermines democratic accountability. Citizens cannot evaluate the safety or efficacy of a system they do not know exists. The absence of external review means that potential biases, error rates, and operational flaws remain unexamined. This lack of scrutiny is particularly concerning given the high-stakes nature of emergency medical dispatch.

Municipal surveillance ordinances typically emerge from public advocacy and legislative response to controversial policing practices. Seattle’s twenty seventeen framework was crafted to ensure that technology deployments align with community values and constitutional protections. By classifying the Corti system as a decision-support tool, the fire department effectively circumvented the legislative intent behind the ordinance. This interpretation creates a dangerous precedent where powerful monitoring capabilities can be deployed under alternative classifications. The absence of public debate leaves citizens without recourse to challenge the technology.

Legal scholars emphasize that the impact on individual rights should dictate regulatory classification rather than technical architecture. When an algorithm directly influences whether a citizen receives emergency medical care, the stakes transcend traditional surveillance definitions. The right to informed consent becomes particularly relevant in healthcare contexts where patients rely on transparent decision-making processes. Municipal governments must establish clear guidelines that prioritize public welfare over operational convenience. The current regulatory gap leaves vulnerable populations exposed to unvetted automated systems.

How does the technology influence dispatcher decision-making and patient outcomes?

The deployment of automated routing prompts has demonstrably altered call distribution patterns within the city. A fire department medical director previously claimed that nurse line routing increased by fifty percent following the system implementation. A subsequent department spokesperson corrected that figure to thirty-two percent, though neither statistic has undergone independent verification. The discrepancy between internal claims highlights the absence of external oversight mechanisms. Even a modest thirty-two percent shift represents a substantial change in how emergency resources are allocated across the metropolitan area.

The consequences of diverted calls can be severe, as illustrated by the case of Pamela Hogan. The seventy-one-year-old woman called emergency services in April of twenty twenty two reporting knee pain. She was routed to the nurse consultation line and waited more than ten hours for a callback. Weeks later, she was discovered deceased in her home. Her estate successfully cleared a legal dismissal hurdle, demonstrating that diverted calls can carry fatal risks. Although her case predates the live AI prompting, it underscores the inherent dangers of redirecting medical distress away from immediate emergency response.

Dispatchers operate under intense pressure to manage limited ambulance resources while ensuring rapid response times. The introduction of algorithmic suggestions adds a new variable to an already complex decision-making environment. While the department asserts that operators can and do override AI prompts, the sustained increase in nurse line routing suggests that the system is highly influential. The psychological weight of an automated recommendation may subtly bias human operators toward following the suggested path. This dynamic raises questions about the true extent of human agency in an increasingly automated dispatch ecosystem.

The psychological dynamics of automated recommendations warrant careful examination. Dispatchers operate in high-stress environments where split-second decisions determine patient survival. An algorithmic prompt can subtly shift operator focus toward the suggested routing path, even when override capability exists. Research in human-computer interaction consistently shows that automated suggestions influence human judgment through confirmation bias and authority deference. The thirty-two percent increase in nurse line routing likely reflects this cognitive bias rather than pure operational necessity.

Emergency medical dispatch protocols have evolved significantly over the past three decades. Standardized assessment tools were originally designed to compensate for human variability and ensure consistent care levels. The introduction of artificial intelligence adds another layer of complexity to an already intricate system. While automation promises efficiency, it also introduces new failure modes that require rigorous testing. The lack of independent verification for routing statistics underscores the need for external auditing mechanisms. Public safety systems demand higher standards of accountability than commercial software deployments.

What are the broader implications for municipal AI governance and oversight?

The Seattle deployment reflects a national trend of municipal governments adopting artificial intelligence without establishing robust regulatory frameworks. National attention regarding AI oversight has primarily focused on frontier language models and national security applications. Local emergency services have received considerably less scrutiny despite directly impacting vulnerable citizens during critical moments. The gap between technological capability and regulatory readiness creates significant legal and ethical vulnerabilities. Municipal leaders must develop governance structures that address the unique risks of automated decision-making in public safety.

Mayor Katie Wilson has acknowledged the need for structured oversight by developing a public-facing AI governance framework. Her administration previously paused the expansion of Microsoft Copilot tools within city government, signaling a recognition that automated systems require careful evaluation. The Corti deployment, however, proceeded for months before any formal governance framework was announced. There is no indication that the fire department sought mayoral approval or conducted a comprehensive risk assessment prior to implementation. This timeline suggests that technological adoption outpaced institutional policy development.

Other Washington counties have adopted similar technologies but with important operational distinctions. Snohomish County utilizes a different artificial intelligence system called Ava to manage non-emergency calls, having processed over two hundred twenty thousand interactions. Kitsap County has also deployed the Aurelian system on a dedicated non-emergency line. These regional deployments deliberately limit algorithmic intervention to lower-priority communications. Seattle’s approach diverges by applying real-time analysis to all medical 911 calls, including life-threatening emergencies that require immediate human judgment.

The unresolved legal questions surrounding the Seattle system will likely be addressed by the city council or the judicial system. Whether the current surveillance ordinance adequately covers real-time call analysis remains an open question. Municipal governments must establish clear standards for transparency, accountability, and public consultation before deploying automated tools in emergency services. The eighteen-month period of undisclosed operation demonstrates how quickly technology can outpace democratic oversight. Future deployments will require proactive governance rather than reactive policy development.

The national landscape of municipal AI adoption reveals a consistent pattern of rapid deployment followed by delayed regulation. Local governments frequently lack the technical expertise to evaluate vendor claims or assess algorithmic bias. This capability gap forces municipalities to rely on vendor assurances rather than independent analysis. The Seattle case demonstrates how quickly operational decisions can outpace institutional capacity. Municipal leaders must invest in technical literacy and establish dedicated oversight committees before approving new technologies.

International comparisons offer valuable insights into regulatory approaches. European municipalities have implemented strict data protection frameworks that require algorithmic impact assessments before deployment. These frameworks emphasize transparency, data minimization, and continuous monitoring of system performance. American cities have largely adopted a voluntary compliance model that prioritizes innovation over precaution. The divergent approaches highlight a fundamental philosophical difference regarding the role of government in technology governance. Seattle’s silence on the Corti deployment contrasts sharply with proactive regulatory environments elsewhere.

Conclusion

The integration of artificial intelligence into emergency dispatch represents a fundamental shift in public safety operations. Municipalities must balance operational efficiency with the ethical obligations of transparency and accountability. The Seattle case illustrates the risks of deploying powerful monitoring systems without public consultation or regulatory review. As automated decision-making becomes more prevalent in healthcare and emergency services, institutional frameworks must evolve to protect civil liberties. Future policy development should prioritize clear definitions, mandatory impact assessments, and continuous public oversight. The technology will continue to advance, but democratic accountability must remain the foundation of its deployment.