Google's selfie video is a recovery recording stored on the account
Google says a signed-in user can store a reference selfie video and later match a new one. It is encrypted at rest. Workspace, child, and Advanced Protection accounts are excluded.
The short version
Google says a person who is already signed in can record a reference video, then later record a new one that is compared with it, including a movement check. The video is encrypted at rest and can be deleted. Workspace accounts, child accounts, and Advanced Protection accounts cannot add one. Using the video to improve Google's face systems is an optional toggle, not the default purpose Google describes.
A reference clip and a later clip
Google's launch post says a person who can already get into an account may record a short selfie video, with guided head movement, and store it for a later lockout. Getting back in means recording another video. Google says it compares the new clip with the stored one and asks for movement so a still photo or a fake video is a weaker attack. The post says the stored video is encrypted at rest, that the person can delete it, and that the purpose is sign-in unless the person opts into something further. A later Google post adds that enrollment happens in a trusted session that is already signed in, and that deleting the clip requires a fresh sign-in, in the same family as other sensitive account changes. The help page says Workspace accounts, child accounts, and Advanced Protection accounts cannot add the video, and that the feature is not on every region, account, or device. Privacy isPrivacy.
What Proton is pointing at
Proton's headline says the selfie video puts a face on Google's servers. That is a location claim. Google's post is a handling claim: encrypted at rest, consent, deletion, sign-in purpose unless you opt in. A video can be encrypted and still be on someone else's server. This brief will not pick a slogan. It will keep the location and the handling rule in the same paragraph so a reader sees they answer different questions. A key that never leaves the holder is a different design.the key stays with the holder.
The optional improvement toggle
The account help page says that when a person submits a selfie video, they can allow Google to use the video and related data to improve facial recognition, age estimation, and other checks that use a body or a movement. The control is reversible. Turning it on is a second purpose, beyond the recovery comparison. This brief will not describe the toggle as required, and it will not describe the recovery flow as if it already included model training. The paper isFutureweb.
Who is outside the offer
Workspace, children, and Advanced Protection are the exclusions the help page prints. A company administrator who expected this on a work account has the page's answer. A personal account that has not been offered the button yet is in the rollout sentence, not in the exclusion sentence. Those are different reasons for a missing setting. Security is Asked directly, Where does the reference video live? Google's launch post says the video is stored with the account, encrypted at rest, and used to help the person sign in unless they opt in to other uses. Proton's headline states the consequence in the other direction: the face is on Google's servers. Both sentences can be true.Security.
Liveness is a check, not a guarantee
Google says it matches the new video to the stored one and requires a simple movement, and that its ordinary sign-in systems can still demand more checks when a session looks wrong. That is a layered claim. It is not a proof that every synthetic video will fail. This page will not rate the matcher. It will say what Google says the matcher is asked to do, and stop. Software isSoftware.
Deletion is an account action
The help page says a person can delete the stored video from the account, and that Google may keep it for a period as a security measure, and longer if it is enforcing a policy violation. Delete on the screen and gone from every disk are different. The page itself inserts the delay. A reader who needs the stricter reading should read that sentence, not a paraphrase that omits it. Hardware isHardware.
What this page will not call the video
This page will not call the clip a passkey, a password, or a local-only secret. Google's own post compares it with passkeys and recovery contacts as other options, which is how a company says the new option sits beside them. Proton's objection is about custody. Both belong in the record. The dependency is who holds the face template after the person has proven they are signed in today. Networking isNetworking.
A sign-in video is not a payments selfie
Google's own community guidance separates this recovery video from selfie checks used for payments products. This brief is about the sign-in feature in the launch post and the account help page. A merchant identity check is a different flow even if a camera is involved. Mobile is Asked directly, Who can add a selfie video? Google's help text says Workspace accounts, child accounts, and accounts in the Advanced Protection Program cannot add one. Everyone else still depends on the feature having reached the account, the region, and the device. The help page says it is not available for every region, account, or device.Mobile.
Where the description sits
The launch post and the later security post are Google's account of the mechanism. The help page is the exclusion list, the toggle, and the deletion delay. Proton is the custody objection. Nothing here is a rating of error rates. The short answer this page will stand on is this: Google says a person who is already signed in can record a reference video, then later record a new one that is compared with it, including a movement check. The video is encrypted at rest and can be deleted. Workspace accounts, child accounts, and Advanced Protection accounts cannot add one. Using the video to improve Google's face systems is an optional toggle, not the default purpose Google describes.
Sources
The reports this brief is filing. Futureweb did not republish them.
Questions
Who can add a selfie video?
Google's help text says Workspace accounts, child accounts, and accounts in the Advanced Protection Program cannot add one. Everyone else still depends on the feature having reached the account, the region, and the device. The help page says it is not available for every region, account, or device.
Where does the reference video live?
Google's launch post says the video is stored with the account, encrypted at rest, and used to help the person sign in unless they opt in to other uses. Proton's headline states the consequence in the other direction: the face is on Google's servers. Both sentences can be true.
What does the later blog say the match is?
A follow-up Google post says the system compares a new live video with the stored reference, and that sign-in can ask for movement so the clip is live. It says adding a video requires an already signed-in, trusted session, and deleting one requires signing in again.
Is improvement of face recognition required?
The help page says a person can allow Google to use the video to improve facial recognition, age estimation, and related checks. The toggle is labeled optional. Leaving it off is the path that keeps the video on the sign-in purpose the launch post describes.
Does this replace a passkey?
Google's launch post says the option sits alongside passkeys and recovery contacts. It does not say the video retires those methods. A holder who wants the secret to stay in a device key still has that older dependency, which is a different page.