Gunshot detection technology has been pitched as a solution to one of the most urgent challenges in public safety: getting officers to the scene of a shooting faster. The premise is straightforward. Deploy microphones across a neighborhood, use AI to identify gunshots, and alert responding units before a 911 call even comes in.
In practice, the results have been far more complicated.
A recent analysis from the Center for Democracy and Technology takes a hard look at how these systems have actually performed in the field. The findings are difficult to ignore. In Chicago, over 50,000 alerts over a 17-month period led to evidence of gun-related crimes less than 10% of the time. In New York City, only 13% of alerts resulted in confirmed shootings. In Dayton, officers filed a crime report for just 5% of system alerts. The NYPD alone wasted over 426 officer-hours in a single month responding to alerts that turned out to be unfounded.
Cities including Chicago, San Antonio, and Charlotte have since stopped using the technology altogether.
The real problem isn’t the idea — it’s the implementation
Faster response times and better situational awareness are legitimate goals. The ambition behind gunshot detection technology is not wrong. The problem is a system that generates noise at a rate that overwhelms the teams it is supposed to support.
When a technology produces false alarms the vast majority of the time, it doesn’t enhance human decision-making. It undermines it. Officers responding to unfounded alerts aren’t just wasting time. They are making decisions in the field based on AI signals that are more likely to be wrong than right. That erodes trust in the technology, strains resources, and in some cases has contributed to wrongful stops and flawed prosecutions.
The CDT analysis also raises a critical point about where these systems tend to be deployed, with significant concentration in majority-minority communities. When AI recommendations carry inherent bias in deployment, the harm isn’t distributed equally.
What good AI-assisted security actually looks like
The lesson here isn’t that AI doesn’t belong in public safety. It’s that accuracy, transparency, and human oversight are not optional features. They are the foundation.
The strongest deployments of AI in security are those where the technology supports human decision-making with clear context and structured workflows, rather than replacing judgment with an automated alert that demands immediate action. Effective systems reduce the noise, not amplify it. They surface what matters, explain why it matters, and give the people responsible for responding the information they need to act with confidence.
That means building systems that learn the specific environments they operate in, that are transparent about what they detect and how, and that are designed from the ground up to keep humans in control of consequential decisions.
Gunshot detection technology, as broadly deployed today, largely fails those tests. But the conversation it has prompted is a useful one. As AI becomes more embedded in public safety infrastructure, the question every agency and operator should be asking is not just whether a system can detect something. It’s whether it can be trusted to do so accurately, consistently, and in a way that makes the people using it more effective rather than more reactive.
Read the full CDT analysis: AI In Policing: Gunshot Detection Technology
Learn how Koshee builds AI video intelligence around human decision-making and schedule a demo