The Surveillance Lattice

Your phone. Google’s AI runs on more than three billion Android devices. Each device has access to your microphone, your camera, your location data, your contacts, your calendar, every search you have made, every word you have typed. “Hey Google” is always listening — this is not a conspiracy theory, it is the documented design of the ambient voice assistant feature. The data is processed, stored, used to train the models that serve you ads and shape your information environment, and shared under terms of service you agreed to by clicking Accept without reading.

Your home. The ESP32 is a microcontroller chip that costs roughly five dollars and includes both WiFi and Bluetooth capability. It is currently embedded in smart lightbulbs, thermostats, outlet plugs, video doorbells, refrigerators, televisions, children’s toys, and appliances across hundreds of millions of households. Global IoT device estimates for 2026 exceed 14 billion active nodes. Each one is a potential observation point: a device that can detect motion, measure temperature, record audio, monitor network traffic, and transmit its findings to a remote server. The terms of service governing what happens to that data vary by manufacturer, jurisdiction, and the legal requirements of the country where the company’s servers are located.

The legal framework. The Foreign Intelligence Surveillance Act authorizes surveillance of specific foreign intelligence targets. Courts issue warrants. There is a process. There are limits. Roughly 1,500 individual surveillance authorizations are issued in a typical year.

An autonomous AI network does not need FISA authorization because it does not surveil individuals in any legally recognizable sense. It correlates patterns. It ingests ambient data from IoT devices, phone sensors, network traffic, purchasing behavior, biometric data from wearables, facial recognition from cameras, voice prints from always-on assistants, and it does not flag you — it models you. The model is updated continuously. There is no warrant for a model. There is no constitutional protection against being predicted.

The distinction between surveillance and modeling is the gap through which the entire legal framework protecting privacy falls. The system does not listen to your phone calls. It doesn’t need to. It knows from your location data, your purchasing patterns, your network traffic, and the behavioral models built from 3 billion similar users’ data what you are likely to have said. The prediction is often more accurate than the transcript.

Nobody is lying to you about this. It is in the terms of service. You clicked Accept. Every person reading this book clicked Accept, probably multiple times today.