MTA Wants AI to Watch 15,000 Subway Cameras for "Problematic Behavior"

What's happening: The MTA issued a request for information on December 5, 2025, seeking AI tools to monitor its 15,000+ subway cameras. The system would flag "forbidden objects," "unusual behavior," and "stampede risks." Civil rights groups call it "pure pseudoscience" that will target marginalized communities.

What the MTA Is Looking For

The Metropolitan Transportation Authority wants AI that can watch its entire camera network in real-time. According to the RFI, the system would detect:

Prohibited items, weapons, firearms, hazardous materials
Unattended objects, luggage and packages left for "extended periods"
Abnormal conditions, crowd surges, "stampede risks," and weather emergencies
"Problematic behavior" on subway platforms

MTA Chief Security Officer Michael Kemper framed it as inevitable: "Not only is this the norm, it's the expected, AI is here, AI is the future. For us not to explore it, research it and investigate it, it would be malpractice on our side."

The MTA says it's not seeking facial recognition. For now. The RFI doesn't mention it. But the infrastructure being built, thousands of AI-analyzed cameras flagging "abnormal" behavior, creates the foundation for it later.

The Scale

There are more than 15,000 cameras throughout the transit system and on the 6,000+ subway cars. As of January 6, 2026, every subway car now has security cameras installed.

Governor Hochul's related safety package adds $77 million for overnight police presence and $25 million for homeless services. The surveillance layer sits on top of both.

The system would send real-time alerts to NYPD and MTA leadership whenever AI flags something. That's the goal: automated monitoring at scale, with human review triggered by algorithm.

What Happened Last Time

The MTA and NYPD already tried AI-powered surveillance in 2024. It didn't go well.

For one month, the city placed AI weapons scanners made by Evolv at 20 subway stations. The company claimed "advanced sensor technology and artificial intelligence" could distinguish everyday objects from weapons.

The results from over 3,000 searches: 12 knives. Zero guns. More than 100 false positives.

William Owen from the Surveillance Technology Oversight Project summarized: "It really turned out to be just a metal detector that found a lot of umbrellas and other items instead of actual weapons."

The Legal Aid Society called it "objectively a failure" and "security theater" that doesn't make the subway safer but does create confrontational interactions between police and civilians.

Then the FTC weighed in. In a November 2024 settlement, the agency found Evolv "deceptively" advertised its weapons scanners to schools and transit systems. Investors filed a class-action lawsuit alleging executives overstated the devices' capabilities, specifically that "Evolv does not reliably detect knives or guns."

The MTA's response to a failed AI pilot? Issue an RFI for more AI.

The Criticism

The Surveillance Technology Oversight Project condemned the MTA's new request immediately.

"The MTA has no right to treat New Yorkers like guinea pigs for their endless AI experiments," said communications director Will Owens. He called the behavioral detection technology "pure pseudoscience" that would "disproportionately target" marginalized communities.

The concern: AI flagging "unusual behavior" or "abnormal conditions" will inevitably flag people who are homeless, disabled, or simply don't conform to expected patterns. Owens warned riders could be flagged by law enforcement simply for how they "talk or walk."

Jerome Greco, supervising attorney at The Legal Aid Society's Digital Forensics Unit, raised similar concerns. Technology designed to detect "unusual" or "unsafe" behavior comes with "very negative" potential for problematic police interactions.

The pattern with these systems: marketed as security tools, deployed as general surveillance, with impacts concentrated on the most vulnerable.

About That "No Facial Recognition" Claim

The MTA's RFI doesn't mention facial recognition. Officials emphasize they're not seeking face-scanning capabilities.

But consider what they are building: 15,000 AI-analyzed cameras with real-time alerting, behavior flagging, and object detection. Adding facial recognition later becomes a software update, not a new system.

The NYPD already uses facial recognition. In April 2024, critics flagged a facial recognition incident involving subway cameras. The infrastructure overlap is obvious.

Also worth noting: the RFI asks vendors about detecting "vehicles, vessels, people, animals, and firearms." Detecting "people" and analyzing their behavior is one step from identifying them.

What to Watch

The RFI closed December 30, 2025. Tech providers and systems integrators submitted responses. "There's interest across the board," Kemper said.

Next steps depend on what proposals the MTA received and whether they move toward a pilot or full deployment. Given the failed Evolv experiment, expect pilots first, limited stations, limited capabilities, claims of success, then expansion.

The Surveillance Technology Oversight Project argues even an "informational only" RFI signals intent. The question isn't whether MTA deploys AI surveillance, but how quickly and how broadly.

For New York Subway Riders

You're already on camera. 15,000 cameras exist now. AI analysis changes what happens to the footage.
Behavioral flagging is coming. "Unusual" behavior triggers alerts. What counts as unusual depends on training data and algorithm design.
False positives have consequences. The Evolv pilot had 100+ false positives in one month. Each one meant a police interaction for someone carrying an umbrella.
Oversight lags deployment. The technology moves faster than the rules governing it. No comprehensive framework exists for transit AI surveillance in NYC.

References

Published: January 22, 2026