In September 2024 the NYPD finished a thirty-day pilot of AI-powered weapons scanners at twenty subway stations. The machines ran 2,749 scans. They flagged 118 people who turned out to be carrying nothing, found twelve knives — the department declined to say whether any were illegal — and detected zero guns.

Two months later the Federal Trade Commission announced a settlement with Evolv, the company behind those scanners. Under it, Evolv agreed to stop making what the FTC called unsupported claims about its products’ ability to detect weapons using artificial intelligence, and to let K-12 schools out of multi-year contracts if they wanted out.

Here is the strange part: the technology is not fake. Evolv now screens something like four million people a day, and its systems have flagged well over a million weapons since 2021. The problem was never that the scanners do nothing. The problem was the distance between what they do and what buyers were told they do. We covered how the technology itself works in an earlier piece; this one is about the evidence.

What the scanners actually are

Walk-through weapons detectors like Evolv Express are not metal detectors in the traditional sense. They use sensors and software to look for the signature of a weapon-shaped object rather than any metal at all, which is why people can walk through with keys and phones without stopping. That throughput is the entire selling point. A venue that screens forty thousand people cannot run everyone through an airport-style checkpoint.

The trade-off is that a system tuned to let ordinary objects through is also tuned to make judgment calls, and judgment calls have error in both directions. Some harmless objects get flagged. Some dangerous ones do not.

What happened in Utica

In Utica, New York, a student walked into Thomas R. Proctor High School carrying a seven-inch knife. The Evolv system at the door did not flag it. The student later stabbed another student multiple times. The school district had committed to a four-million-dollar contract to put the scanners in thirteen schools; after the incident, it pulled them and went back to conventional metal detectors and X-ray bag screening.

That case, more than any statistic, is what the FTC action was about. The allegation was not that the scanners never work. It was that the marketing described a capability — reliable detection of weapons, including knives, without slowing anyone down — that the product could not consistently deliver.

Why this matters more in New York than most places

New York is the densest concentration of the exact environments these systems are sold into: office lobbies, residential towers, hotels, arenas, schools, hospitals. The city itself ran the highest-profile pilot in the country and published the numbers. And in 2026 Evolv began offering a “Screened by Evolv” seal that venues can put on their doors — which means a building’s security posture is now, in part, a marketing decision made in public.

A seal on the door signals something to visitors. It should mean something to the owner, too, and the owner is the one who will answer for it if it doesn’t.

Five questions to ask before you buy

1. Ask for the detection rate on knives, in writing, from a test you can see. The Utica failure was a blade. Most marketing focuses on firearms. If the vendor cannot show independent test data for edged weapons in the sensitivity setting you will actually run, you are buying a promise.

2. Ask what the false-positive rate does to your lobby. The subway pilot flagged roughly one in twenty-three people. In a building with a thousand entries a day that is forty secondary searches, each one needing a trained person to handle it. That is a staffing cost, not a software feature.

3. Ask who does the secondary screening, and how they are trained. A scanner that flags someone and has no one competent standing behind it has added a delay and nothing else. The machine is the first layer. The person is the one who actually finds the weapon.

4. Ask what the vendor’s claims are backed by since the FTC order. This is not a rhetorical question. The settlement specifically bars unsupported claims about AI detection. A vendor that is still leading with “AI-powered” and no data has not read its own consent order.

5. Ask whether the system is designed for your threat or for a stadium’s. Walk-through detectors were built for high-volume, low-dwell environments. A residential lobby with two hundred residents who all know the doorman has a different problem, and may be better served by a person and a policy than by a machine and a seal.

The point

Weapons detection technology is a legitimate tool, and in the right setting it catches things people miss. But it is a tool with a documented error rate, sold by an industry the FTC has already had to correct once. The scanner at the door is not the security program. It is one input into a program run by people who know what to do when the light turns red — and what to do when it doesn’t.

Worldwide Security Options integrates weapons detection screening into staffed security programs for buildings, venues and schools across New York and New Jersey — with trained personnel behind every scanner, because the machine is only the first layer. Talk to our team about what fits your building.