Evolv AI weapon scanners in NYC subway detect zero guns, 118 false positives in pilot
New York City's month-long pilot of Evolv Technology's AI weapon scanners in subway stations resulted in zero guns detected but 118 false positives that subjected innocent passengers to stop-and-search operations. The system was shelved after results revealed the technology's unreliability for subway environments.
What happened
New York City deployed Evolv Technology's AI-powered weapon scanners in 20 of its 472 subway stations for a month-long pilot beginning at the end of July 2024.
During approximately 2,800 scans, the system detected zero guns and 12 knives, but produced 118 false positives that led to additional stop-and-search operations on innocent passengers.
The pilot was shelved after the NYPD revealed the results to City & State in response to questions about the program's silence.
Reported impact
One hundred eighteen New Yorkers were subjected to additional stop-and-search operations and had their privacy invaded despite carrying no weapons, according to Legal Aid Society attorney Diane Akerman.
The Legal Aid Society characterized the outcome as 'objectively a failure' and noted that the 12 knives detected likely included legal knives for which no arrests were made.
Index analysis
The scanner system suffered from extremely high false positive rates relative to actual detections, identifying non-threats at a rate exceeding actual weapon finds by nearly 10 to 1.
Evolv's CEO told investors in spring 2024 that subways were not a good use case for the technology due to railway interference, raising questions about the appropriateness of the deployment.
Governance controls
City Hall claimed the scanners acted as a crime deterrent because there were no shootings at deployed stations during the month, despite the pilot covering only 20 of 472 stations for a single month, limiting the ability to demonstrate deterrent effects.
The NYPD stated it had not entered into a contract with Evolv Technology and was still evaluating the pilot's outcome.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.