New Orleans Tests Carbyne AI for 911 Call Triage — What It Means for Emergency Tech

New Orleans is testing Carbyne's AI Emergency Call Triage system for 911 calls, according to a report from the Shreveport Times. The Orleans Parish Communication District (OPCD) deployed the AI to handle the high volume of calls — over a thousand per day — by automatically triaging calls related to the same incident.

How It Works

The AI sits in front of human dispatchers. When a call comes in, the system asks if the caller is reporting the currently active incident (e.g., a crash causing a surge of reports). If yes, the AI provides information or updates. If no, the call is transferred to a human dispatcher. This reduces the number of calls that would otherwise pile up during major events.

OPCD states the AI will not be used to handle emergency calls — only to direct them to humans. It's a triage layer, not a replacement for dispatchers.

Context: 311 Experience

New Orleans already implemented AI in April for 311 non-emergency calls. OPCD says 50% of 311 calls are purely informational, so the AI handles those directly. The 911 triage builds on that experience.

Risks and Concerns

The article flags several potential issues with AI 911 triage:

• Hidden biases from training data — particularly if predictive policing algorithms are used, which could reinforce systemic racism via over-policed neighborhood metrics.
• Speech recognition failures for strong accents, dialects, or atypical pitch/articulation, since the AI relies on automatic speech recognition (ASR) trained on limited samples.
• Unreliability without human oversight, thorough training data, strong cybersecurity, and algorithm regulation.

These are not hypothetical: the technology is being tested now in a live 911 environment.

What This Means for Developers

If you're building AI agents for critical infrastructure, this is a real-world case study. The AI isn't replacing human judgment — it's filtering and routing to reduce load. The technical challenges are familiar: natural language understanding with high variability, false positives, and the ethical implications of automated decision-making in emergencies.

The source article doesn't list Carbyne's specific API endpoints or architecture, but the core pattern — classify intent, route to human if ambiguous — can be implemented with any LLM-based classifier + workflow engine.