New Orleans receives more than 1,000 emergency calls every single day. And as of late July, some of those calls — the ones where someone is having a heart attack, or watching a car wrap around a telephone pole, or whispering because her boyfriend is in the next room — are being answered by artificial intelligence.
Not assisted by AI. Answered by it.
The program is called AI Emergency Call Triage, and it's run through the Orleans Parish Communication District. According to the Daily Wire, New Orleans is the first major American city to put AI on the front end of its 911 system, filtering what it considers "urgent" calls from "redundant" or "lower-priority" reports. Karl Fasold, the Executive Director of the Orleans Parish Communication District, says the AI isn't meant to replace people.
That's always what they say right before it replaces people.
The system currently handles two scenarios: when someone reports a crash within 200 meters of an already-reported crash, the AI flags it as a duplicate. And when every human dispatcher is tied up, the AI steps in to field the call. They tested it first on the city's 311 non-emergency line, which is a reasonable enough sandbox. But 311 is "my streetlight is out." 911 is "my husband has a gun."
Those are not the same conversation.
Former 911 dispatcher Boxy, who spent 11 years on the job, called it "the worst idea you could possibly think of." The reason is something anyone who's ever been on a real emergency call understands instinctively: people in crisis don't speak clearly. They slur. They whisper. They use coded language because the threat is standing six feet away. "You need a person," Boxy said. A person who can hear the tremor in a voice and know something is wrong even when the words say everything is fine.
AI doesn't hear tremors. AI hears data inputs.
And the data inputs themselves are a problem. A February report found that AI voice models similar to ChatGPT and Gemini fabricate details nearly 50% of the time — while sounding completely confident doing it. Researchers at China's Zhejiang University demonstrated that voice-based AI models can be externally manipulated with success rates between 79 and 96 percent. So the system answering your emergency call can be fooled by a bad actor almost every time, and when it's not being fooled, it's inventing information on its own about half the time.
Great foundation for life-or-death decision-making.
New Orleans isn't entirely alone in this experiment. Atlanta uses AI to assist dispatchers by pulling up relevant information while a human stays on the line. Seattle has been running a similar rapid-response triage system since 2024, though the city managed to keep that quiet for more than two years before anyone noticed. Which tells you everything about how much public input they wanted on the decision.
The pitch is always efficiency. The city can't hire enough dispatchers, wait times are too long, AI can handle the overflow. And those are real problems. But the solution to "we can't staff our emergency lines" isn't "let's put a chatbot on them." The solution is to pay dispatchers more, or train more of them, or fix whatever bureaucratic mess is keeping the positions unfilled. You don't automate the last line between a person in danger and the help they need because your HR department can't get its act together.
The people who built this system will point to the narrow use cases — just duplicates, just overflow. The people who've watched government technology programs expand know that "narrow use case" is a launch pad, not a fence. Every pilot becomes a program. Every program becomes the default.
A 911 call is someone at their worst moment asking for help from their own government. New Orleans just decided that moment isn't important enough for a human voice on the other end.
