Meta will retain some of its facial recognition systems, including its DeepFace model, and plans to incorporate biometrics into its metaverse products

Vox 2021-11-05 Rebecca Heilweil

Context & Ripple Effects

Meta's decision draws a boundary after it moved to shut down Facebook's decade-old facial-recognition system and delete face-scan data amid legal and regulatory pressure. Rather than abandon the technology wholesale, it is preserving DeepFace and positioning biometrics for metaverse products.

That distinction matters in the later coverage: Meta returned to facial recognition for narrowly framed anti-scam and account-recovery uses, including a test for celebrity-scam ads and compromised accounts and a UK rollout following regulator engagement.

First-order effects

Meta retains DeepFace and related facial-recognition capability even as its consumer Facebook face-scan system is being wound down, preserving a technical base for biometric features in its metaverse products.
Users of Meta's metaverse products become the prospective audience for biometric features, while Facebook users lose the prior platform-wide facial-recognition system and its stored face scans.

Second-order effects

Meta can frame future facial-recognition deployments around defined product functions rather than broad identification, a positioning later visible in opt-in celebrity-scam protection.
Regulators become a practical gatekeeper for Meta's biometric expansion: the later UK test followed engagement with regulators, making deployment scope and consent central to rollout choices.

Third-order effects

The coverage points to facial recognition becoming more compartmentalized at Meta: broad social-platform identification is retrenched while purpose-specific biometric systems are retained for new interfaces and safety workflows.
If that pattern persists, product teams will compete on whether they can tie biometric collection to a bounded benefit—such as account recovery or scam prevention—rather than treat face recognition as a general platform feature.

The trend: Meta is shifting facial recognition from a broad Facebook-era identity feature toward narrower, product-specific biometric uses shaped by regulatory acceptance.

Discussion

@accountabletech @accountabletech on x
Just because Facebook makes a big show of something doesn't mean we should fall for it. Although Facebook has pledged to abandon facial recognition, it's still fair game for other Meta platforms including Instagram and WhatsApp. https://mashable.com/...
@hypervisible @hypervisible on x
Meta-extractive, meta-invasive, never meta data point they didn't like. https://www.buzzfeednews.com/ ... https://twitter.com/...
@katecrawford Dr. Kate Crawford on x
So Facebook is deleting one billion facial recognition scans, but it's keeping DeepFace, the model that is trained on all those faces. Note that “the company has also not ruled out incorporating facial recognition into future products.” Very meta. 👀 https://twitter.com/...
@matthew_d_green Matthew Green on x
Does anyone believe Facebook will really delete those face recognition templates, as in “nobody will keep a copy anywhere” delete?
@notdan @notdan on x
Do you believe @Facebook is actually deleting facial recognition data? [Poll]
@joshuafarkas Josh Farkas on x
Meta lost a $650m privacy lawsuit because of DeepFace in 2020 and $5B FTC fine in 2019, which called out this tech specifically. That's 56% of 2022's investment in the metaverse. The decision to turn off DeepFace was likely risk mitigation, not moral clarity. https://twitter.com/…
@panopticindia @panopticindia on x
We welcome this move however we want to highlight that there are still reasons to be concerned about. 1. DeepFace: FB's facial recognition algorithm that it trained off of the facial data it had still exists. 4/n https://research.fb.com/...
@docmattmoudi @docmattmoudi on x
2. Yes, it deleted/will delete the database used for FR, but as @ellajakubowska1 has said, it retains the software product - DeepFace - that processes faces. Facebook is at liberty to start using this again whenever it decides “societal concerns” have shifted once again. [5/9] ht…
@katerosebee @katerosebee on x
To be clear: still a good thing to not have a gigantic facial identification dataset hanging around to be breached or reused on even worse ideas later. I just am skeptical that FB's decision wasn't largely based on having already extracted value from it.
@ellajakubowska1 Ella Jakubowska on x
#Facebook deleting their facial recognition database? In the words of the great Greta: blah blah blah 😴. They're keeping the algorithm and open to using it in future FB products. I'm finding this hard to see as anything other than a PR move... https://twitter.com/...
@panopticindia @panopticindia on x
2. This move has come as a result of self-regulation in the wake of FB (or Meta) trying to rebrand itself. There is still no law in the US which will regulate FR and thus nothing stopping Meta from going back to using it again even if FB doesn't. 5/n https://twitter.com/...
@ecmclaughlin Elizabeth C. McLaughlin on x
It's called DEEPFACE?!?!? Facebook is just Bond-level villany. https://twitter.com/...
@katerosebee @katerosebee on x
Facebook deleted all that faceprint data because the recognition modeling created, DeepFace, is already trained. They get to look magnanimous by tossing something they don't need anymore.
@ronwyden Ron Wyden on x
Shutting down this program is Facebook doing the bare minimum to protect its users' privacy. Now Congress needs to pass a comprehensive privacy law to crack down on all the other ways that Facebook and shady data brokers abuse Americans' personal information. https://twitter.com/…
@profcarroll David Carroll on x
Told ya right here first: https://www.vox.com/...
@ryan77anderson @ryan77anderson on x
@Grady_Booch @Meta genuine question on GDPR -> re “right to be forgotten ” Q: would it not be impossible to keep using model in Europe, post data flush; because a MODEL REBUILD of ‘verifiably-compliant’ model (Minus any GDPR flagged EU faces) is no longer possible? @katecrawford …
@rik_ferguson Rik Fërgüson on x
... and in a flash, the GDPR “right to erasure” became meaningless. https://twitter.com/...
@tyang209 @tyang209 on x
I am getting rid of my bank heist tools but keeping the money. https://twitter.com/...
@mcclure111 Mcc on x
@katecrawford @adrishaw Hi, could you clarify. The NYT article says Facebook is keeping the *algorithm*. But you say they are keeping the *model*. That sounds like a big difference. To me algorithm means “we're keeping the software” and model means “we're keeping the results of t…
@viss @viss on x
@mcclure111 ... i dont have a source for this, its how im interpreting the various articles. its a very facebook thing to do and seems very on-brand
@viss @viss on x
@mcclure111 ... sounds like theyre ditching the data the model was trained on, and keeping the ‘already trained part’. youre dead on.
@mcclure111 Mcc on x
@Viss ... That's not what the New York Times article says, to my reading. Am I missing something? Can you point out a source saying this (that they're keeping the trained model)?
@mcclure111 Mcc on x
@Viss ... I agree that Facebook lying about this would be on-brand
@shiringhaffary Shirin Ghaffary on x
“While Meta says that facial recognition isn't a feature on Instagram and its Portal devices, the company's new commitment doesn't apply to its metaverse products, Meta spokesperson Jason Grosse told Recode.” https://www.vox.com/...
@nytimes @nytimes on x
Breaking News: Facebook plans to delete face scan data from over 1 billion users, shutting down a facial recognition system that became a privacy headache. https://www.nytimes.com/...
@grady_booch Grady Booch on x
So, @Meta decided their liability regarding facial recognition was higher than its value, threw away the training data in a public announcement, and kept the trained model. Meta is a profoundly unethical company, and it starts at the top, with Mark. https://twitter.com/...
@rmac18 @rmac18 on x
@kashhill Anyone reading today's move should be aware that Meta, Facebook, watever is reserving the right to deploy facial recognition on future products. Its future CTO discussed the possibility of using it in smart glasses in an internal meeting earlier this year. https://twitt…
@profcarroll David Carroll on x
@katecrawford Those cartoon avatars aren't gonna make themselves.
@petermdenton Peter Denton on x
@katecrawford @samfbiddle Once you got the model, you got the model. Do we not remember when FB had to remove Acxiom? It didn't matter! They already had a correlation model between behavior and consumption, based on demographics from Acxiom. Once you know the predictors, you just…

Chronicles