New Orleans deploys AI to screen duplicate 911 crash calls

New Orleans has become one of the first U.S. cities to deploy artificial intelligence in its 911 emergency call system, though the technology operates under strict constraints designed to preserve human oversight.

The Orleans Parish Communication District (OPCD), which handles emergency communications for the city, began using Carbyne's AI-V Emergency Call Triage system in 2023 to manage a specific problem: multiple callers reporting the same traffic accident while human operators are already handling other emergencies.

How the system works

The AI can only activate when two conditions are met simultaneously. First, every human call-taker must be occupied with other calls. Second, an incoming call must originate within roughly 650 feet of a crash that 911 has already documented.

When the AI answers, it identifies itself as New Orleans 911 and asks whether the caller is reporting an accident at a specific location. Any response other than an affirmative answer immediately routes the call back to the human queue. If the caller confirms they're reporting the known crash, the AI informs them that emergency services are already aware and offers the option to stay on the line for a human operator.

The system cannot dispatch responders, determine call priority, or end calls—callers must hang up themselves. According to OPCD, the technology is not a large language model like ChatGPT but a purpose-built tool for this single use case.

Why it matters

Staffing shortages plague 911 centers nationwide, and New Orleans is no exception. A single traffic accident can generate dozens of calls from passing motorists, creating a surge that delays responses to unrelated emergencies. By filtering confirmed duplicate reports during peak periods, the AI aims to free human operators for calls requiring judgment, medical instruction, or immediate dispatch decisions. The approach represents an early test case for how AI might support—rather than replace—emergency communications staff in resource-constrained systems.

Safeguards and performance

OPCD trained the system using local 911 recordings to recognize New Orleans accents and street names. The AI supports multiple languages and defaults to a human operator when it cannot understand a caller or when a call falls outside its narrow parameters.

During its first three months of operation, OPCD reviewed 100 percent of AI-handled calls and reported finding no false positives or false negatives—meaning the system correctly identified duplicate crash reports without misrouting other emergencies. The agency continues to audit the technology and has not expanded it beyond this single use case.

Carbyne, the system's vendor, reported that during a 90-day period the AI triaged more than 3,500 events and reduced redundant calls by over 30 percent, with distressed callers reaching human operators up to 40 seconds faster on average. Because these figures come from the vendor rather than an independent audit, ongoing transparency from OPCD will be essential.

What comes next

OPCD is testing other potential applications in a controlled environment but has not deployed AI for any additional 911 call categories. The agency has stated it will not expand the system unless it can ensure the technology performs reliably.

All AI interactions are recorded with standard metadata and audio, and OPCD says call data is not uploaded to external large language models. If a caller hangs up while waiting, standard callback procedures remain in effect.

These details were first reported by Kurt Knutsson for Fox News.