The world of automated surveillance is booming, with new machine learning techniques giving CCTV cameras the ability to spot troubling behavior without human supervision. And sooner or later, this tech will be coming to a store near you — as illustrated by a new AI security cam built by Japanese telecom giant NTT East and startup Earth Eyes Corp.

The security camera is called the “AI Guardman” and is designed to help shop owners in Japan spot potential shoplifters. It uses open source technology developed by Carnegie Mellon University to scan live video streams and estimate the poses of any bodies it can see. (Think of it like a Kinect camera, but using 2D instead of 3D data to track people.) The system then tries to match this pose data to predefined ‘suspicious’ behavior. If it sees something noteworthy, it alerts shopkeepers via a connected app.

A demo video of a prototype version of the technology published by Earth Eyes gives a good overview of how it works:

AI Guardman has been under development for at least a few years, but last month NTT East and Earth Eyes shared the results of some early trials with the camera. As you might expect for a PR blast, the feedback was glowing; and according to a report from Japan’s IT Media, NTT East and Earth Eyes claim that AI Guardman reduced shoplifting in stores by around 40 percent.

Without independent verification, these claims should be taken with a pinch of salt, but the underlying technology is certainly solid. New deep learning techniques have enabled us to analyze video footage more quickly and cheaply than ever before, and a larger number of companies in Japan, America, and China are developing products with similar capabilities. Similar features are also making their way into home security cameras, with companies like Amazon and Nest offering rudimentary AI analysis, like spotting the difference between pets and people.

AI Guardman is notable, though, as a product with advance features that customers will be able to buy, plug in, and start running without too much delay. A spokesperson for NTT East told The Verge that the camera would go on sale at the end of July, with an up-front price of around $2,150 and a monthly subscription fee of $40 for cloud support. NTT says it hopes to introduce the camera to 10,000 stores in the next three years. “Our primary target is big businesses although we do not have the intention to omit small ones,” said a spokesperson.

But there are a lot of potential problems with automated surveillance, including privacy, accuracy, and discrimination. Although AI can reliably look at a person and map out where their body parts are, it’s challenging to match that information to ‘suspicious’ behavior, which tends to be context dependent. NTT East admitted as much, and said that “common errors” by the AI Guardman included misidentifying both indecisive customers (who might pick up an item, put it back, then pick it up again) and salesclerks restocking shelves as potential shoplifters.

It’s difficult to know the extent of this problem, as NTT East said it had not published any studies on the technology’s accuracy, and could not share any information on statistics like its false positive rate (that is, how often it identifies innocent behavior as something suspicious).

It’s also possible that the training data might be biased towards certain groups, or that the technology might be used as a pretext for discrimination. (I.e., a security guard following someone around the store because ‘the computer said they’re suspicious.’) NTT East denied that the technology could be discriminative as it “does not find pre-registered individuals.”

Evaluating technology like this is difficult at a distance, but it’s clear that this sort of automated surveillance is only going to become more common in the future, with researchers working on advanced analysis like spotting violent behavior in crowds, and tech companies selling tools like facial recognition to law enforcement. Next time you walk past a CCTV camera, your concern won’t be who is watching, but what.