"Off" Doesn't Mean Off: What We Found Testing AI Wearable Power States
During routine testing at the AFR lab, a secure, controlled environment for security and AI research, we observed something that challenged a basic assumption most people hold about their devices: turning something “off” should mean it stops.
The observation
We were testing WIRESHIELD’s AI recording device detection capabilities against several wearable devices. After completing a test session, a researcher powered down one of the AI wearables through its companion app. The app confirmed the device was off. The LED went dark. By every visible indicator, the device had shut down.
WIRESHIELD disagreed.
On the next scan, hours after the device was “powered off”, WIRESHIELD flagged the wearable as present in the environment. The Bluetooth Low Energy radio was still actively broadcasting advertisement packets. The device was still announcing its presence to anything listening.
Why this happens
This isn’t a malfunction. It’s by design.
Bluetooth Low Energy wearables maintain their radio in a low-power advertising state to preserve the ability to reconnect with their paired phone. When you tap “power off” in a companion app, the app typically instructs the device to:
- Stop recording or processing audio
- Disable the LED or visual indicators
- Enter a low-power state
What it does not do is disable the Bluetooth radio. The radio stays alive because the device needs it to:
- Reconnect when you open the app again (without going through the pairing process)
- Receive commands like firmware updates or configuration changes
- Maintain its bond with the paired smartphone
This is standard behavior across nearly all BLE wearables, not just AI recorders. Fitness trackers, smartwatches, and wireless earbuds all do the same thing. The difference is that an AI recording wearable broadcasting its presence has different privacy implications than a fitness band.
What “off” actually means
We tested multiple power states and found a consistent pattern:
| State | Bluetooth Radio | Detectable? |
|---|---|---|
| Active (recording/processing) | Broadcasting | Yes |
| “Off” via companion app | Broadcasting | Yes |
| Unpaired / factory reset | Broadcasting (discoverable mode) | Yes |
| Battery depleted | Silent | No |
| Physical power switch (if available) | Silent | No |
The only reliable way to silence a BLE wearable’s radio is to exhaust its battery or use a physical power switch, which most AI recording wearables don’t have.
The privacy implication
Consider a meeting where a colleague has an AI recording wearable. They tell the room, “Don’t worry, I turned it off.” They genuinely believe it’s off, the app says so, and there’s no visible indicator. But the device is still broadcasting. Anyone with a BLE scanner can confirm it’s present.
Whether the device is still capturing audio in this state depends entirely on the manufacturer’s firmware implementation. The companion app may have told the device to stop recording, but there’s no way for anyone in the room to independently verify that. The Bluetooth signal simply confirms the hardware is still active.
This gap between perceived state and actual state is the kind of ambiguity that erodes trust in shared spaces.
What you can do
If you carry an AI wearable:
- Understand that “off” in the app means “reduced functionality,” not “powered down”
- If privacy is critical, physically remove the device from the environment, don’t rely on software power states
- Be transparent with people around you about what the device does and doesn’t do when “off”
If you’re concerned about nearby devices:
- Use wireless signal analysis tools to verify what’s actually broadcasting in your environment
- A device showing as “off” to its owner may still be detectable via Bluetooth
- The presence of a Bluetooth signal doesn’t confirm recording is active, but it confirms the hardware is not powered down
Our approach
WIRESHIELD detects AI recording wearables based on their Bluetooth signal characteristics, regardless of what the companion app reports as the device’s power state. If the radio is broadcasting, we flag it. We don’t make assumptions about whether the device is recording, we surface the fact that it’s present and active at the hardware level.
This is part of our broader philosophy: security intelligence should be based on observable signals, not reported states.
This research was conducted at the AFR lab, a secure and controlled environment operated by Andromeda Field Research LLC for security and AI research purposes. All testing was performed on devices owned by researchers in a controlled setting.