Public Safety — New Orleans puts AI agents on 911 calls
New Orleans is letting an AI agent answer 911 calls, offloading redundant incident reports from understaffed human dispatchers — and doing it without telling callers they're talking to a machine. It's a preview of the transparency fights coming to every understaffed public call center.
New Orleans' Orleans Parish Communication District (OPCD) is testing Carbyne's AI Emergency Call Triage, which now answers 911 calls and routes non-urgent reports away from human dispatchers. The system is wired into the city's computer-aided dispatch: callers within 200 meters of a live auto accident get routed to an AI agent that asks whether they're calling about the crash, and only transfers to a human if the answer is no — a pattern OPCD found was burying critical calls during rush-hour surges. The district, which handles over a thousand emergency calls a day and estimates it is about 40% short of full staffing, says AI triage recoups at least two hours of call-taker time a day. Carbyne's case study claims 3,500+ events triaged over 90 days and a 30% drop in redundant calls.
Why it matters: this is the highest-stakes AI deployment in the country, and there are no rules for it. Callers are not told they're speaking to an AI, no city policy or national framework governs AI in emergency response, and the failure modes are documented. During the Bourbon Street attack investigation, an AI agent took the first two 911 calls and briefly classified the incident as a hit-and-run before a human took over — exactly the kind of error that matters when seconds count. The district says the AI will not handle emergencies, only direct calls to a human, but the boundary is exactly where triage systems have failed before: the system decides what counts as an emergency.
The context is a city that's been here before: OPCD deployed AI on its 311 non-emergency line in April, where roughly half of calls are information requests. 911 is a different category entirely — it's life-critical, and the stakes of a misclassification are measurable in response time, not convenience. New Orleans is one of the highest-volume call centers in the country, with about 18 million annual visitors, so this is a real-world stress test of whether AI triage can hold up at scale.
The governance gap is the story. There's no requirement to disclose AI to callers, no agreed standard for when AI must hand off to a human, and no national framework — each understaffed district that adopts this will make up its own rules. The useful debate isn't whether AI should triage 911 (the queue is genuinely broken and this is a sensible fix); it's whether the public has a right to know they're talking to a machine, and who is accountable when the machine's classification is wrong.
What to watch: whether other understaffed 911 centers follow New Orleans, and whether any jurisdiction moves first on caller-disclosure rules.
If you called 911 and a machine answered, should it have to tell you? Tell us in the comments.
Sources: Shreveport Times · GovTech · Carbyne · Hacker News discussion
