New Orleans has become the first major U.S. city to implement artificial intelligence in answering 911 calls, a move that has drawn both praise for potential efficiency and sharp criticism over safety risks.
The Orleans Parish Communication District (OPCD) is now routing emergency calls to an AI agent that determines which situations require human dispatchers and which can be handled by the bot. This system, named AI Emergency Call Triage, was introduced in late July after a trial period on the city's 311 hotline.
According to OPCD Executive Director Karl Fasold, the AI is not intended to replace human dispatchers but rather to assist when all human operators are busy and a caller is located within approximately 200 meters of a reported crash. The AI is designed to handle repeat or lower-priority calls, aiming to prevent dispatchers from being overwhelmed.
However, critics, including individuals identifying as first responders, have voiced strong opposition, describing the system as potentially dangerous. Concerns center on the AI's ability to accurately interpret calls from individuals who may be distressed, have accents, or are in situations where they cannot speak clearly, such as domestic violence incidents where callers might feign other reasons for their call.
These critics argue that human dispatchers are essential for discerning nuances in a caller's voice and situation that an AI might miss, emphasizing the need for human judgment in life-or-death scenarios.
While New Orleans is the first major municipality to use AI in this capacity, other cities have adopted AI in their 911 systems to varying degrees. Atlanta utilizes AI to help live dispatchers pinpoint caller locations more rapidly. Seattle has employed similar technology since 2024 to aid dispatchers in prioritizing emergency responses, a practice that reportedly went undisclosed to the public for over two years.
Concerns about AI's reliability in voice communication are supported by recent studies. Research indicated that AI models, including prominent ones like ChatGPT and Gemini, have fabricated information nearly half the time, presenting it with a convincing audio delivery. The already high-stress and often unclear nature of 911 calls could exacerbate the risk of AI misinterpretation.
Furthermore, AI voice systems are known to be vulnerable to external attacks and can develop biases based on the data they are trained on. Cybersecurity researchers have demonstrated the ability to manipulate voice models' behavior with success rates ranging from 79% to 96% in some tests, and external hacking attempts have also proven effective in causing these systems to malfunction.