Google's Smart Glasses Can Translate Speech on the Fly, and Try to Protect the People Around You

Google has shown its Android XR software running on smart glasses and mixed-reality headsets, with its Gemini AI assistant built in. Android XR is the operating system that runs these devices, which blend digital content with the real world. In a hands-on test, Google's software ran on glasses made by Samsung and sold under the Warby Parker and Gentle Monster names. The Verge published its hands-on report on November 12, 2026.

The most useful feature is live translation. To start it, the wearer presses two fingers to their temple. The user does not pick languages or download a language pair first. Most offline translation tools need that setup step, much like packing a phrasebook before a trip.

Privacy was the main focus of the testing. The glasses automatically redact, or hide, faces in photos and video so that people nearby are not identifiable. Gemini declined to identify a person in view or in a photo, citing privacy protections.

The Warby Parker and Gentle Monster models have a privacy light that turns on when the device intends to save visual data. Using the camera and microphone requires permission that the user must opt in to. Google said it will filter out background noise, including conversations of bystanders. Its beam-forming microphones, which focus on sound from one direction, will process only the wearer's audio. The Gentle Monster pair includes a 12-megapixel (12MP) camera.

Google also described what happens to data after capture. It said faces are redacted before glasses data reaches its servers. The company said it will not use that data to target ads, will not use it to train Gemini, and that humans will not use it to improve its models. Google also plans to disable the glasses if someone tampers with the privacy safeguards. That is a stated plan.

In my view, the questions that matter most are practical, not promises on paper. Blurring faces before upload, a light tied to recording, and a pledge to keep data out of ads and AI training are all meaningful design choices. Each one depends on details that a demonstration cannot confirm. Those include how well face detection works in crowds or dim light, how the light behaves if someone tampers with it, and how permissions are set up for businesses. Developers and security teams will need those answers before drawing firm conclusions.

The broader context here is where the assistant lives. A translator that needs no language setup has to send sound or context to an AI model to work. That is why privacy is built into the product from the start. Google places its privacy controls at the moments of capture and upload. Whether those controls hold up as the glasses spread is the thing to watch.

The demonstrations covered three areas: controls on the device when recording, rules for handling data on the server, and a translation feature that asks very little of the wearer.

In my view, treating bystander privacy as a default is an important choice for a product category that has struggled to earn public trust with cameras worn on the face. The next stage will show whether those safeguards hold up in everyday use.