A Samsung TV running the Tizen operating system appears to grab a 960x540 image of your screen every couple of seconds and feed it into a TensorFlow Lite model called alex.tflite, according to a technical breakdown shared on X by a researcher who goes by mike. The model is an AlexNet implementation, the researcher says, and its job is to guess which game is currently on screen.
The claims come from a live inspection of a Tizen device rather than an official Samsung statement, so treat them as findings and leaks, not confirmed policy. The same researcher describes the project as a broad look at Tizen concerns, with more detail still to be published. The story matters because it turns a vague privacy worry into a concrete, testable fact: a known neural network file, a known image size, and a known refresh rate.
Samsung Tizen screen tracking: alex.tflite specs
The most striking part is how ordinary the tooling is. A standard pre-trained network, shrunk to 8-bit integers to run cheaply on a television, is used to answer a very simple question: what is this screen showing? Keeping the imagery at 960x540 suggests the goal is classification, not detailed inspection of small text or personal documents.
The researcher is blunt about the results. The nearest-neighbor matching, he says, does not work in about 99 percent of cases, and the output is dumped without being acted on because the consent check happens later in the flow.
- Model: alex.tflite, an AlexNet-style network following torchvision.models.alexnet.
- Input: 960x540 screen captures, running with INT8 quantization every couple of seconds.
- Method: nearest-neighbor feature comparisons against a database of reference images.
