We love to pay contractors for excrement that doesn't work in the name of security theater
When Mayor Eric Adams announced last week that the City would soon be testing AI-powered weapons-detection systems in the subway, and posed next to one such system, made by a company called Evolv, he touted the feedback the system has gotten from other customers. "What I'm hearing from my corporations, from my hospitals, from others, they're saying this is living up to our expectation," he said.
But Hell Gate contributor Felipe De La Hoz got his hands on the numbers showing what happened when the Evolv system was tested at Jacobi Medical Center in the Bronx in 2022: The system threw up a huge number of false alarms, triggering on roughly one in four people to pass through it, even though less than half of one percent of those alarms actually constituted an actual threat.
NYC Has Tried AI Weapons Scanners Before. The Result: Tons of False Positives
By Felipe De La Hoz
Last week, Mayor Eric Adams announced that he was kicking off a 90-day waiting period to test weapons scanners on New York City's subway system.
The scanners that he demonstrated, provided by Evolv Technology, purport to mix physical detection technology with artificial intelligence to indicate when someone is carrying a gun or a knife, and have been used in City hospitals before, something Adams alluded to at the press conference at Fulton Street station.
"We are going to make sure, based on our numbers, hit ratio, false hit, we're going to do our data," Adams told reporters. "What I'm hearing from my corporations, from my hospitals, from others, they're saying this is living up to our expectation."
What Adams didn't mention was that the City already has some data, and it's not good. Evolv's scanners were installed at Jacobi Medical Center in the Bronx in 2022, and triggered huge numbers of false positives, according to the results of the pilot obtained by Hell Gate through a public records request filed with NYC Health + Hospitals.
Over the seven months that the Jacobi pilot was active, 194,000 people passed through Jacobi's scanners, and in just over 50,000 of those cases, the scanners threw up an alarm—an incidence rate of around 26 percent, or over one of every four times someone passed through the scanner. Of those 50,000 alarms, around 43,800, or a little more than 85 percent, were false positives; 7,027 of the alarms, or 14 percent, were law enforcement officers who were presumably carrying their service weapons; and just 295 alarms, or 0.57 percent, were determined to be a non-law enforcement person carrying either a knife, a gun, or a threat type labeled only "other," which likely entails other weapons like bats.
Notably, Evolv's scanners did not get any more accurate as the pilot progressed—there was not a single month where the alarm to visitor ratio fell below 25 percent. In September of 2022, the final month of the seven-month pilot, 27,900 visitors passed through the scanners, and nearly 7,000 threw alarms; out of those 7,000 alarms, just 345 potential threats were identified—a false positive rate of 95 percent, with only 0.45 percent of alerts being non-law enforcement threats. Throughout the entire pilot, the alerts led to the finding of 24 guns, 139 knives, and 132 other potential threats, out of the 50,000 alarms sounded.