TL;DR: The New York MTA put out a call for AI companies to help monitor 15,000+ cameras across 472 subway stations. They want algorithms that detect weapons, unattended items, and "unsafe behaviors." Civil rights groups are furious. The Surveillance Technology Oversight Project calls it "pseudoscience" that will disproportionately flag disabled riders and people of color. Last year's AI weapon scanner pilot in NYC found zero guns, 12 knives, and over 100 false positives. Now they want to scale it citywide.
The MTA's AI Wish List
The MTA isn't hiding what it wants. In December 2025, the transit authority issued a formal Request for Information (RFI) to tech companies, asking what AI capabilities they could offer [1].
The ask is broad:
- Detect weapons in real time
- Monitor unattended items
- "Anticipate subway stampedes"
- Flag "unsafe behaviors"
- Create a "proactive intelligence-driven ecosystem"
That last phrase is key. The MTA admits their current monitoring is "manual, reactive and resource intensive." They want algorithms running 24/7 across 15,000 cameras in stations and 6,000 more in train cars [1].
MTA Chief Security Officer Michael Kemper defended the move: "For us not to explore it, research it and investigate it, it would be malpractice" [2].
Malpractice? That's an interesting choice of words for an agency about to conduct behavioral surveillance on 4 million daily riders.
NYC's AI Track Record: Failure After Failure
The MTA acts like AI surveillance is some exciting frontier. But New York has already tried this. It didn't go well.
The Evolv Gun Scanner Disaster (2024)
Remember when Mayor Eric Adams and the NYPD put AI-powered weapon scanners in subway stations? Here's how that went:
- 3,000+ searches across 20 stations
- Zero guns found
- 12 knives recovered
- Over 100 false positives [3]
The Legal Aid Society called it "objectively a failure." The FTC later found that Evolv "deceptively" advertised its technology [4].
William Owen from the Surveillance Technology Oversight Project put it bluntly: "It really turned out to be just a metal detector that found a lot of umbrellas" [2].
Behavioral AI Testing (2024)
The MTA has already experimented with AI systems that score whether riders are behaving "rationally." Yes, seriously. An algorithm deciding if you're acting normal enough [4].
Fare Evasion AI (2023)
The MTA contracted with a company called Awaait to use AI to monitor riders for fare evasion. The message is clear: surveillance comes first, whether it works or not [4].
Civil Rights Groups: "Pseudoscience"
The Surveillance Technology Oversight Project (STOP) didn't hold back in their condemnation:
"The MTA has no right to treat New Yorkers like guinea pigs for their endless AI experiments." [4]
Will Owen, STOP's communications director, called "behavioral detection" AI what it is:
"Any AI software that claims to flag so-called unusual behavior is pure pseudoscience, disproportionately targeting [Black, indigenous and people of color] and disabled transit riders." [4]
Think about who moves "unusually" in a subway: people with disabilities, autistic individuals, those with mental health conditions, elderly riders, tourists unfamiliar with the system. These are the people an algorithm would flag.
STOP added: "New Yorkers should never have to worry they'll be flagged by law enforcement simply for the way they talk or walk" [4].
The Surveillance Numbers
Let's put the scale in perspective:
15,000+ Station Cameras
Across 472 subway stations throughout the five boroughs
6,000+ Train Cameras
Every car in the subway fleet now has security cameras
4 Million Daily Riders
All potentially subject to AI behavioral analysis
24/7 Operation
The system never stops running
The MTA claims this RFI is "informational only." But STOP argues it signals clear intent to expand surveillance citywide [4].
What About Facial Recognition?
Notably, the MTA's request doesn't mention facial recognition technology. But that's cold comfort.
The NYPD already uses facial recognition extensively, including in an April 2024 subway incident that drew criticism from investigators [2]. The infrastructure being built for "behavioral AI" can easily be repurposed for facial recognition later.
First they build the cameras. Then they add the AI. Then they connect the databases. The surveillance state grows one "reasonable" step at a time.
The Accountability Game
Jerome Greco, from the Legal Aid Society's Digital Forensics Unit, raised a crucial point:
"These uses of AI are not like Netflix telling you what movie you should watch next. The consequences of it being wrong could be pretty significant." [2]
When Netflix's algorithm is wrong, you get a bad movie recommendation. When behavioral AI is wrong in a subway, someone could be detained, questioned, humiliated, or worse.
The MTA has provided no timeline for implementation. An unspecified number of tech companies responded to the December 30 deadline. The next step involves reviewing proposals [2].
There's still time to push back.
What You Can Do
Contact the MTA Board
The MTA board holds public meetings. Demand transparency about AI surveillance plans. Ask what safeguards exist against bias. The board's contact info is on the MTA website.
Support STOP
The Surveillance Technology Oversight Project is fighting this. Follow them at stopspying.org for updates and ways to take action.
Document Everything
If you're approached or questioned in the subway, write down what happened. Contact the NYCLU or Legal Aid Society if you believe you were targeted unfairly.
Know Your Rights
You can refuse to answer questions beyond identifying yourself. You don't have to consent to searches. If detained, say: "Am I free to go?" and "I want a lawyer."
References
- THE CITY: MTA Explores How to Use AI to Monitor Thousands of Cameras in Transit System (January 8, 2026)
- StateScoop: AI cameras will flag 'problematic behavior' in New York subways, MTA says (January 2026)
- Planetizen: New York MTA Wants to Use AI to Detect Weapons and Other Threats on Transit System (January 2026)
- StateScoop: Civil rights group 'condemns' NYC transit authority's pursuit of AI video analytics systems (January 2026)