MIT is installing more than 500 AI-powered surveillance cameras across its campus, and the price tag tops $3 million. According to Hacker News, which surfaced reporting from The Tech, the cameras are going into academic buildings, residence halls, and outdoor areas along Memorial Drive. Installation started in November 2025 and is expected to run through September 2026.
This isn’t a few security cams over a loading dock. It’s a campus-wide network built to watch, classify, and remember.
What the cameras actually do
The hardware comes from Hanwha’s Wisenet AI line, monitored by AI software called Ai-RGUS. These aren’t passive recorders. Per the technical specs cited by Hacker News, the system can collect real-time face and object classification data, including:
- Motion, loitering, and crowd detection
- Face mask and camera-tampering detection
- Automatic classification of people by clothing color, gender, and age
- Recognition of faces, license plates, and vehicles in real time
The cameras can classify individuals from up to 35 feet (11 meters) away, support resolutions from 2MP to 4K, and most can pan, tilt, rotate, and zoom. MIT says collected data is “retained up to 30 days,” unless an exception is granted.
Why this matters
What stands out here is the shift from recording to interpreting. Older campus cameras just stored footage that someone reviewed after an incident. These run deep learning models continuously, tagging people and objects as they move. That’s a different capability, and it changes the relationship between an institution and the people inside it.
MIT is one of the most influential research universities in AI. When a place that helps build this technology deploys it at scale on its own students and staff, it sets a reference point. Other universities and large campuses watch what MIT does. If this becomes the template, expect it to spread.
The automatic classification piece is the part worth slowing down on. Sorting people by age, gender, and clothing color isn’t the same as reading a license plate. It’s demographic profiling at a distance, done by algorithms that carry known accuracy gaps across skin tones, ages, and lighting conditions. A system that’s wrong even a small percentage of the time, run across 500 cameras and thousands of daily passersby, produces a lot of wrong calls.
The bigger pattern
This fits a broader trend. AI vision models got cheap and good enough that surveillance vendors now bundle classification into off-the-shelf camera lines. What used to require a dedicated engineering team is now a product you buy and mount. Hanwha’s Wisenet line is marketed exactly this way: point it at a scene, and it identifies and sorts objects out of the box.
That lowered cost is the real story for practitioners. The barrier to deploying analytic surveillance has collapsed. Budget approval, not technical capability, is now the main gate. Expect more institutions to cross it.
What to watch next
A few things are worth tracking as this rolls out:
- Policy pushback. Faculty, students, and privacy groups tend to respond once these systems go live and people feel watched. Retention rules and classification features are the likely flashpoints.
- Scope creep. A 30-day retention window and “unless an exception is granted” language leaves room to widen over time. Where the exceptions land will tell you how the system is really used.
- The precedent. If MIT normalizes campus-wide AI classification, other schools and corporate campuses will cite it. This is the kind of deployment that quietly becomes a standard.
For anyone building or buying AI vision tools, MIT’s move is a marker of where the technology sits right now: capable, affordable, and being installed faster than the rules around it are written. The infrastructure is going up through September 2026, so this story has a long runway. You can find the full details at the original source.