Google’s new smart glasses remove bystanders’ faces from images before its Gemini AI processes them, the company said in a blog post on November 12. Google describes this protection only for images sent to Gemini when a wearer asks about what they are seeing. The post does not say whether it also applies to photos and videos that wearers save to their gallery.

“When technology can see the world around you, it has a responsibility to the people around you, too,” Shahram Izadi, Google’s vice president and general manager for XR, wrote in the post.

Google built the glasses with Samsung, while Warby Parker and Gentle Monster designed the frames. They go on sale in the US and South Korea on December 10, starting at $395. Wearers can say “Hey Gemini” to ask about what they are looking at or press a button to take photos and videos.

What does Google promise?

• Face removal: The glasses are “designed to automatically redact bystanders’ faces” before Gemini processes an image.
• No storage or training: Gemini does not save camera images in the user’s history, and Google does not use them to train its AI or target ads. However, Google can still use the text of Gemini chats to improve its AI models unless users turn this off.
• Wearer’s voice: Gemini’s microphones focus on the wearer’s voice.
• Recording light: A light turns on while the glasses take photos or videos. The glasses stop recording if the light is covered.
• Usage rules: Wearers must agree not to record people without permission or in sensitive places. Google may disable the glasses of those who break these rules.
• App rules: Google Play bars apps from using the camera for real-time facial identification or live-streaming, or from changing how the light behaves. Google will not advertise products that hide the light.
• Other safeguards: Data on the glasses is encrypted, a physical switch turns them off, and they are available only to users aged 18 and above.

These promises clearly cover images that the glasses send to Gemini. To understand how they apply to other recordings made using the glasses, we examined five common situations.

Scenario 1: Asking Gemini about a street mural

A wearer looks at a mural and asks, “Hey Gemini, who painted this?” To answer, the glasses capture an image of the mural, which also shows people walking past.

• What Google says: The glasses remove the faces of passers-by before Gemini sees the image. Gemini does not save the image, and Google does not use it to train its AI.

Scenario 2: Recording friends at a café

A wearer records a video of friends at a café. Other customers are sitting in the background, and some of them are talking.

• What Google says: The recording light tells people nearby that the wearer is filming.
• What the blog does not say about faces: Google says it removes faces only from images sent to Gemini. The blog does not say whether the faces of other customers remain visible in videos saved to the gallery.
• What the blog does not say about audio: The launch blog says videos capture spatial audio, which records sounds from all around the wearer. Google says its microphones focus on the wearer’s voice when they talk to Gemini, but it does not say whether this also applies to video recordings. If it does not, the videos could capture conversations among nearby customers.

Scenario 3: Recording someone without permission

A wearer films a stranger on the metro without asking for permission or records patients in a hospital waiting room.

• What Google says: Wearers must agree not to record people without permission or in sensitive places. Google may disable the glasses of those who break these rules.
• What the blog does not say about enforcement: The blog does not explain how Google finds out when a wearer breaks these rules, given that recordings remain on the wearer’s device.
• What the blog does not say about complaints: It does not explain whether strangers or patients can report such recordings to Google, especially if they do not use Google’s products.

Scenario 4: Recording a threat and asking Gemini for help

A stranger threatens a wearer on the street, and the wearer records the incident as evidence. Later, the wearer asks Gemini to help draft a police complaint using that video. For the recording to serve as evidence, it needs to show the stranger’s face.

• What Google says: The glasses remove faces from images sent to Gemini, and Gemini does not save those images.
• What the blog does not say: It does not clarify whether these protections also apply when a wearer shares a previously saved recording with Gemini.

Scenario 5: Using another app

A wearer opens a translation app on the glasses to read a menu in a busy restaurant. The app uses the camera, so the captured image also shows other diners.

• What Google says: Apps can use the camera only with the wearer’s permission. Google Play bars them from using it for real-time facial identification or live-streaming.
• What the blog does not say: It does not clarify whether the glasses remove diners’ faces before an image reaches a third-party app, as they do before sending images to Gemini.

What we asked Google:

MediaNama sent Google the following questions on November 13. We will update this story when it responds.

1. Does face removal apply when a wearer uploads a saved photo or video from the glasses to Gemini later, for example through the phone app? Does Google’s promise not to save camera images or use them for training cover such uploads?
2. Does Google remove bystanders’ faces from photos and videos that wearers save to their gallery? If not, why not?
3. Do third-party apps with access to the glasses’ camera receive images with bystanders’ faces removed?
4. Do videos record conversations of people around the wearer? Does the focus on the wearer’s voice apply only to Gemini, or to video recordings too?
5. Does face removal run on the glasses or in the cloud? How accurately does it detect faces, for example in low light, crowds or at an angle?
6. How does Google find out when a wearer records people without permission or in sensitive places? Has it disabled any wearer’s glasses so far?
7. Can people who don’t use Google’s products report a recording made on the glasses? If so, through what channel, and what does Google do with such reports?
8. Does Google plan to launch the glasses in India? If so, how will these protections apply under India’s Digital Personal Data Protection (DPDP) Act?

What this means for India:

Google has not announced an India launch yet. Meta’s Ray-Ban glasses, however, have been on sale in India since May 2025, and people in India have already used them to record bystanders. Most of these incidents began as ordinary recordings that wearers later posted online. Our question to Google about gallery recordings addresses this concern.

• Café in Khan Market: A creator covertly recorded a man at a Delhi café on July 24 and posted the video as a reel, which crossed 4 lakh views. Instagram said the video did not violate its Community Standards. The Internet Freedom Foundation (IFF) sent a legal notice to Meta and the creator on August 24, seeking Rs 2.05 crore. The notice argues that Meta is liable as the maker of the glasses and that its safe harbour as an intermediary does not cover this claim.
• Street harassment: A man followed a woman down a dimly lit street on June 11 and recorded her using Meta’s glasses. He posted the clip from an account that sells paid “coaching” on approaching women, and it received about 23 lakh views. Police arrested him, and IFF sent a second notice on September 10. The woman does not have an Instagram account, so she could not use Meta’s grievance process.
• Trans rights protest: A creator recorded a transfeminine artist at a protest at Jantar Mantar on March 29, and the clip spread as a meme. Instagram initially said the video did not violate its rules but removed it only after weeks of campaigning by LGBTQIA+ groups.
• Police use: Delhi Police used Meta’s glasses to record demonstrators at a protest in June. At another protest at Jantar Mantar on October 10, Indian Express Hindi reporters spotted an officer wearing the glasses. The officer said he was not recording.

In July, Meta updated its glasses to turn off the camera if someone tampers with the recording light. Instagram head Adam Mosseri said the platform would take down videos that harass people. However, as of October, sellers on Amazon India still listed stickers that block the light.

India’s DPDP Act does not apply to personal data that individuals process for personal or domestic purposes, and its core provisions will take effect only in May 2027. No Indian court has decided whether a glasses maker is liable when its device records a bystander. Until a court settles this question, bystanders depend largely on safeguards such as those Google describes.

Also Read: