New York City's subway system is moving toward a new layer of automated surveillance. The Metropolitan Transportation Authority (MTA) plans to use AI software to watch for suspicious activity on subway platforms and send alerts to police.
The proposal sits at the intersection of public safety, machine learning, surveillance cameras, and civil liberties. It follows a series of attacks on the subway, while also drawing criticism from groups that say the system may be too broad and too error-prone.
What the MTA Plans to Do
The MTA's plan centers on software that reviews live feeds from surveillance cameras on subway platforms. Instead of relying only on people watching screens, the system would automatically flag behavior it interprets as suspicious and notify police.
MTA head of security Michael Kemper described the goal as "predictive prevention." In practical terms, that means the agency wants to identify potential danger before it turns into an incident, using camera feeds already present across the transit system.
The plan does not include facial recognition, according to MTA spokesperson Aaron Donovan. That distinction matters because it separates this system from tools designed to identify specific people by matching faces against databases.
Even without facial recognition, the system still represents a major expansion in how video can be used. The question is not only whether cameras exist, but what the software is trained to notice, how often it is wrong, and what happens after an alert reaches police.
A Camera Network Already Covers the Subway
The MTA has installed surveillance cameras on every subway platform and inside every train car. About 40 percent of those cameras are monitored in real time.
That existing network gives the agency a large base for AI monitoring. If software is connected to live feeds, it can process activity at a scale that would be difficult for human staff alone to match.
For riders, the visible change may be limited at first. The cameras are already there. The more important shift is behind the scenes: video that was once mainly watched or reviewed by people can become part of an automated alert system.
This creates a different kind of surveillance environment. A camera records what happens; an AI system also makes a judgment about whether what it sees should trigger attention from law enforcement.
Why Supporters See a Safety Tool
The MTA's push comes after a series of attacks on the subway. In that context, automated monitoring is being framed as a way to respond faster to threats on platforms.
Subway platforms are busy, crowded, and constantly changing spaces. A system that can scan live camera feeds may help surface incidents that human monitors miss, especially when only about 40 percent of cameras are watched in real time.
The core promise is speed. If software can detect possible danger and alert police quickly, the MTA may be able to intervene earlier than it otherwise could.
But that promise depends on the system's accuracy. A tool designed for prevention must still distinguish between genuinely risky situations and ordinary behavior that looks unusual only to software.
Why Civil Liberties Groups Object
Civil liberties groups, including the NYCLU, have criticized the plan as excessive. Their concern is not only that cameras are present, but that AI will add another layer of automated judgment to public transit surveillance.
NYCLU policy counsel Justin Harrison warned that AI systems are prone to mistakes and could worsen existing inequalities. That warning points to a central risk in automated policing tools: when a system gets something wrong, the consequences may fall unevenly on different riders.
A false alert is not a minor issue if it leads to police attention. In a subway setting, the person flagged may not know why they were stopped or what behavior caused the system to react.
The MTA's statement that the software will not use facial recognition addresses one category of concern, but it does not end the debate. A system can avoid identifying faces and still raise questions about surveillance, error rates, accountability, and how police respond to alerts.
The Bigger Question for Subway AI
The New York City subway AI plan shows how public agencies are beginning to use machine learning not just to review information, but to shape real-time decisions. In this case, the decision is whether activity on a platform should be treated as suspicious enough to alert police.
That makes transparency important. Riders may want to know what kinds of behavior the system flags, how alerts are reviewed, and what safeguards exist when the software is wrong.
For the MTA, the challenge is to show that the technology improves safety without becoming an excessive surveillance tool. For critics, the concern is that automated monitoring could normalize a broader form of policing in everyday public spaces.
The plan is therefore about more than one AI deployment. It is a test of how cities balance prevention, public trust, and civil liberties when software begins watching live transit spaces alongside human operators.