← Back to home

Known smart glasses BLE patterns

Unrecorded uses a local, open-source catalogue of hardware fingerprints — not machine learning. When visible Bluetooth fields may match a known pattern, the app can raise a calm privacy-risk indicator. A match means possible proximity to compatible hardware, not proof of recording.

Important: Name keywords, address-prefix hints, and manufacturer IDs are weak, independent signals. They can produce false positives and miss devices that hide their identity. Treat every alert as situational awareness.

How pattern matching works

For each observation from a scan batch, Unrecorded checks:

  1. Benign filter first — common headphones, TVs, keyboards, and fitness bands are down-ranked so they are less likely to trigger wearable alerts.
  2. Catalogue match — advertised names, cautious address-prefix hints, and documented manufacturer IDs are compared to entries below.
  3. Session merge — repeated sightings within an in-memory ScanSession can modestly increase confidence before stale signals expire.
  4. Risk score — low / medium / high indicators with plain-English reasons. Never certainty.

Source of truth in the repo: detection_signatures.dart.

Brand families in the catalogue

The table summarises what the open-source engine looks for. Exact keyword lists and hints may change between app releases.

Brand family Example name signals Supporting hints
Meta / Ray-Ban ray-ban, meta glasses, stories Address-prefix hints; manufacturer ID 0x0D53 (weak)
Snap / Spectacles spectacles, snap glasses Address-prefix hints; manufacturer ID 0x03C2 (weak)
Even Realities even realities Name keyword
Focals focals Name keyword
Vuzix vuzix Name keyword
Xreal xreal, nreal Name keyword
INMO inmo Name keyword
TCL RayNeo rayneo, tcl rayneo Name keyword
Solos solos Name keyword
Generic smart glasses smart glasses Broad phrase — higher false-positive risk
Wearable camera (generic) camera glasses, wearable camera Generic — may not be smart glasses specifically

What we deliberately filter out

To reduce noisy alerts, name patterns associated with common benign devices are checked before catalogue matching. Examples include AirPods, Galaxy Buds, Bose/JBL speakers, Roku/Chromecast TVs, Fitbit, and Garmin watches. A nearby headphone should usually be classified as audio hardware — unlikely recording — rather than a wearable camera.

Why a match is not identity proof

Read the full limits in our detection limitations FAQ.

Testing patterns without real glasses

Enable Demo Mode in app settings, or broadcast a test advertisement from another phone using a tool like nRF Connect with a custom name such as ray-ban or meta. This validates the alert path — not real-world certainty.