Gun detection is about to become a line item in six Chicago police districts, and the only public scorecard for the technology is a ratio nobody at City Hall wants to read aloud: 82 alerts, roughly 10 firearms, six arrests.

The pilot is meant to begin by January 1, 2027 and run six months in the 2nd, 5th, 7th, 9th, 10th and 11th districts, the South and West Side beats where the city’s shootings actually concentrate. Drones, license plate readers and assorted other artificial intelligence tools are supposed to be tested alongside it, bundled into one trial, and four companies are apparently lined up to run the whole thing. The city will not name the four (CBS News Chicago reported). The police department’s command staff, by the account of the alders who asked, was not consulted on what would be tested, only on where. That is not a procurement process. That is a rumor with a budget line.

ZeroEyes, the AI gun detection vendor already running at Navy Pier, is one of the names in circulation. Its chief strategy officer has described talks with the Johnson administration as fairly advanced. Talks are not an award, a scope of work, or a definition of success. Chicago alders spent a Monday night budget hearing trying to extract those things and got a description of an intention instead.

ZeroEyes — 82 Alerts Is Not a Track Record, It Is a False-Positive Rate

The nearest thing Chicago owns to evidence sits on the transit system. The CTA put ZeroEyes gun detection on 250 train-platform cameras in a $200,000 trial starting in the summer of 2024, a program the agency announced in August 2024. By April 2025 the agency’s own security leadership was reporting the numbers to its board: about 82 detections across the run, roughly 10 of them judged to be actual lethal firearms, and six arrests (the CTA board record). The board then approved a $1.2 million extension pushing coverage toward 1,500 cameras through the summer of 2026, over objections from civil liberties advocates about the absence of public input (WBEZ and the Sun-Times).

Do the arithmetic the way a risk officer would. Eighty-two alerts to produce ten guns means something close to seven of every eight flags were not guns. A drill on a platform. A phone held at the wrong angle. A hand inside a jacket. Each one of those is a human at a control center deciding, in under two minutes, whether to push an armed response to a public station at rush hour.

ZeroEyes’ answer is that a person, a former service member or cop, reviews the frame before anything reaches a dispatcher. The model casts a wide net and a human filters it, and the company sells that review as the thing separating it from a dumb motion alert. As engineering, that is a reasonable design. As accountability, it means the product is a queue, and the city is buying a vendor’s headcount decisions without being shown the error rate on the other end of them.

Gun Detection on the CTA — Ten Guns, Six Arrests, and a Train It Did Not See

Here is the sentence that belongs taped inside the procurement office. In September 2024, four people asleep on a Blue Line train were shot to death. The gun detection system did not alert police (WBEZ and the Sun-Times). It was pointed at those platforms. It was live. It had nothing to report, because the weapon was not brandished in a camera’s line of sight before the killing started.

That is not a malfunction and no vendor should apologize for it. It is the design’s entire blind spot. The model looks for a visible firearm held in a way a training set has seen before. It does not hear a shot, does not read intent, does not notice the man walking down the car with his hand in his coat. A camera that only sees the moment of display will always be a camera that sees everything except the part where someone could have been stopped.

The camera saw the train. The train was never the problem.

Ald. Ray Lopez — A Pilot Nobody Can Describe Is a Purchase Order Waiting to Happen

The politics are uglier than the technology. Chicago voters backed restarting gunshot detection by more than 72% in an advisory referendum, a mandate built almost entirely on frustration rather than evidence. Into that gap walked an administration promising a no-cost test, with no vendor list, no published metrics, and a schedule that had already slipped once. Ald. Ray Lopez called the whole presentation a ruse to get through the election and said he would not be surprised if the selected company had been set up to fail. Ald. Bill Conway said a no-cost program of this scale is hard to believe. Ald. Samantha Nugent and Ald. Anthony Beale pressed for a straight answer and did not get one (the hearing).

The eventual permanent gunshot detection contract is described as a multimillion-dollar award, separate from the pilot and coming after it. Read the sequence carefully, because it is the oldest move in municipal procurement: a vendor absorbs six months of cost, the hardware footprint stays bolted to the light poles, the public gets used to the alerts, and by the time the evaluation memo is due, the alternative is no longer a competing product. It is a city that has to explain why it switched the cameras off.

Sharla Roberts — “No Cost” Is a Sentence That Requires a Signature

Chicago’s chief procurement officer told the council her office is still building the pathway to test these tools. Still building it, seven weeks before the announced start date, with four vendors unnamed and a police department that was not asked what it wanted tested. A six-month evaluation with no success criteria registered in advance is not an experiment. It is a runway with a landing light on the far end.

No-cost pilots are also never free. The cost moves: into the permanent contract that follows, into the officer hours spent answering alerts that turn out to be a wrench, into the civil liability of a stop made on a bad frame at two in the morning. Somebody in the 9th or the 11th district pays that, and it is not the vendor.

Axel’s Take — The People in Those Six Districts Are the Test Set

Every automated system that points at a neighborhood has an error rate, and the error rate is always paid in the same currency: a person standing on a sidewalk with his hands where an officer tells him to put them. That is the honest description of what Chicago is about to run in six districts. The city is not testing whether gun detection works in the abstract. It is testing how many wrong alerts the residents of the South and West Sides will absorb before someone decides the math is not worth it.

The districts chosen are the ones that already hosted ShotSpotter and then lost it after years of disputed results, and they are the ones where a real homicide clears at a rate nobody would call satisfactory. Those neighborhoods keep getting the first experiment and the last audit. If the city were serious, it would publish the alert count, the confirmed-firearm count and the arrest count monthly, in public, per district, before the pilot begins. Nothing in the current plan requires that.

There is also a comparison nobody at that hearing was asked to make. Chicago does not face a choice between gun detection and nothing. It faces a choice between a vendor’s camera model, a vendor’s acoustic sensors, and a few hundred more detectives who can work a case after the shot instead of a control center that can only narrate it. The council was denied the information to make that call, and it is being asked to bless the pilot anyway.

Chicago is not buying safety here. It is buying the appearance of a system, and the invoice arrives after the pilot closes, when the cameras are already up and the only remaining question is who signs.

Sources: CBS News Chicago, WBEZ, Citizen Portal, PR Newswire.