First published November 3, 2021.
On November 2, 2021, Facebook's parent company Meta said it would switch off the Face Recognition system on Facebook and delete more than a billion facial recognition templates. Anyone who had opted in would stop being identified automatically in photos and videos, and the feature that suggested tagging friends would disappear.
The short version for a reader who wants to know what changed: the photos stayed, the face-matching stopped. A "template" is a numerical summary of a face that software uses to match one picture against others. Deleting templates meant Facebook could no longer say "this is you" in a new upload, but it did not remove a single photo from anyone's albums.
What Meta actually announced
The announcement came in a post by Jerome Pesenti, who was then Meta's vice president of artificial intelligence. It listed several concrete changes that rolled out over the following weeks:
- Automatic recognition of people in photos, videos and Memories ended for everyone who had switched the setting on.
- Suggested tags and name suggestions were removed, along with the opt-in setting itself.
- Notifications telling you that you might appear in someone else's post, based on your face, stopped.
- Automatic Alt Text, the tool that describes images for blind and low-vision users, kept describing pictures and counting people, but no longer named them.
Meta said more than a third of Facebook's daily active users had chosen to turn on face recognition, which explains why the number of templates passed one billion. Pesenti called the move "one of the largest shifts in facial recognition usage in the technology's history." Wikipedia's account of the system adds that deletion was planned for December 2021.
What the decision did not cover
The announcement was narrower than many headlines suggested. Meta did not say it was giving up on face recognition as a technology. The same post described identity checks, fraud prevention and unlocking a locked account as useful cases, and said the company would keep exploring them, including systems that run on a person's own device so the face data never leaves it.
It also did not mean that the underlying research vanished. Facebook's DeepFace model, described in a 2014 research paper, had powered the tagging features since about 2015. According to reporting summarised on Wikipedia, the company deleted the user templates but did not plan to destroy the DeepFace software itself.
So the fair summary is this: one of the largest consumer face-identification systems in the world was turned off, and the faces it had learned were deleted, but the door to narrower, consent-based uses was left open.
Why Facebook stepped back
Meta framed the decision as a balance between benefits and "growing concerns about the use of this technology as a whole," and it noted that regulators had yet to set clear rules. Pesenti wrote that the technology's long-term role "needs to be debated in the open." Behind that careful language sat several years of legal and political pressure.
Lawsuits and regulators
In Illinois, the Biometric Information Privacy Act requires companies to get written consent before collecting biometric identifiers such as face geometry. A class action over Facebook's photo-tagging feature ended in a $650 million settlement, approved by a federal judge in early 2021, with about 1.6 million Illinois residents eligible for payments of at least $345 each.
At the federal level, the US Federal Trade Commission's July 2019 settlement with Facebook, which carried a record $5 billion penalty, required the company to give clear notice about how it used face recognition and to get users' affirmative consent before using it in ways that went beyond what it had previously disclosed.
Public trust
The timing also mattered. By late 2021 Facebook was facing intense scrutiny over leaked internal documents and had just renamed its parent company Meta. Giving up a controversial feature was an easy way to show users and lawmakers that the company could limit itself.
How face recognition works, and where it fails
A face recognition system usually works in steps. It detects a face in an image, measures features such as the distance between the eyes or the shape of the jaw, turns those measurements into a template, and compares that template with others in a database. In "one-to-one" mode it checks whether you are who you claim to be, as when a phone unlocks. In "one-to-many" mode it searches a large gallery for a match, as police systems do.
The second mode is where most of the controversy lies. In December 2019 the US National Institute of Standards and Technology (NIST) tested 189 algorithms from 99 developers. It found that, for many of them, false matches were 10 to 100 times more likely for Asian and African American faces than for white faces in one-to-one checks, and that African American women had higher false-match rates in one-to-many searches. NIST also found that the most accurate systems tended to be the most even-handed, so the problem is real but not identical across products.
False matches have real consequences when the technology is used by police. Several people in the United States have reported being wrongly arrested after a face search pointed at them, which is why some cities restricted police use of the technology.
Common misunderstandings about the shutdown
"Facebook deleted a billion users' photos." No. It deleted face templates, not images. Your pictures and the tags you had already added stayed where they were.
"Facebook sold everyone's face data." There is no evidence of that. Criticism of Facebook's data practices often mixes different issues. The Cambridge Analytica scandal, for example, involved a third-party app harvesting profile data from millions of users, not the sale of face templates. The face recognition concern was mainly about consent and future misuse.
"Meta abandoned face recognition for good." Also not quite. The 2021 post explicitly kept the option of narrower uses. Later reporting said Meta began testing face matching again in 2024 to detect scam adverts using celebrities' images and to help people recover locked accounts, this time as a limited, opt-in feature.
What it means for your own privacy
If you used Facebook in 2021, there is nothing you need to do about the old templates. They were scheduled for deletion by the company. Your photos remain, and so do the tags people added by hand, which can still connect your name to an image.
A few habits help on any platform:
- Review tagging settings so you can approve tags before they appear on your profile.
- Check which apps have access to your photo library on your phone and your social accounts.
- Read any prompt that asks you to turn on face features: the choice to opt in is the main safeguard most laws rely on.
- Remember that public photos can be collected by third parties. Companies such as Clearview AI built search tools from images scraped from the open web, and have faced lawsuits in the US and regulatory action elsewhere as a result.
Facebook's 2021 decision did not settle the debate over face recognition, but it showed that a company with every technical reason to keep the feature found the legal and reputational costs too high. That is a useful signal for anyone weighing whether the convenience of automatic tagging is worth the data it requires.
Sources: Meta: An Update On Our Use of Face Recognition; Wikipedia: DeepFace; NIST: study on race, age and sex in face recognition software; FTC: $5 billion penalty and new privacy restrictions on Facebook; Wikipedia: Facial recognition system; earlier coverage by Futurism.
COMMENTS