Apple can’t escape trial in massive class action over photo face scans
(Legal Newsline) — A federal judge has largely rejected Apple’s push to avoid either a trial or a settlement payout in a potentially massive class action seeking damages under Illinois’ biometrics privacy law.
In an opinion filed Sept. 30 — which had been sealed until Oct. 8 to allow for possible redactions — U.S. District Judge Nancy Rosenstengel again refused to scuttle litigation accusing Apple of allegedly illegally scanning and identifying the faces of millions of people who were imaged in photos uploaded through iPhones and other devices to the company’s Photos app.
The ruling follower her June 5 opinion granting plaintiffs’ request to certify the lawsuit as a class action, essentially expanding the legal action under Illinois’ Biometric Information Privacy Act to encompass perhaps as many as 6.5 million additional class member plaintiffs.
“Plaintiffs have framed one of the common questions as whether Apple collected biometric data from each class member and used it to identify the class member,” Rosenstengel wrote in June. “Plaintiffs have presented evidence that the Photos app scans each image for potential faces and collects faceprints, which are feature vectors that uniquely identify face images but do not associate a name with the face. …
“Specifically, the app deploys ‘machine learning algorithms’ on the device to recognize people from their visual appearance. It is also undisputed that Apple knows the identities of its device users, including their names and email addresses, which are associated with their Apple IDs.”
In arguing for summary judgment, which would have ended the case in their favor without trial, Apple insisted it neither collects nor holds any biometric information or identifiers.
The plaintiffs class also moved for partial summary judgment on the issue of corporate liability.
As a first matter, Rosenstengel denied Apple’s request to incorporate into the record additional material facts, as it said the plaintiffs didn’t explicitly dispute the evidence. But she said the plaintiffs did have a chance to address the facts in response to the summary judgment motion and noted their argument “that 96 of the 123 additional material facts are identical to facts that they already disputed or admitted, while the remaining 27 were previously addressed in substance.”
Turning to the BIPA claims, Rosenstengel reiterated there was no dispute about whether Apple complied with rules about data retention and destruction or followed written consent guidelines. She said the question is whether Apple’s face crops and faceprints constitute “scans of face geometry” and whether those files are capable of determining identities.
Crops, she said, are not biometric identifiers. She explained Apple software runs every picture in a user’s Photos app through a machine learning algorithm that detects faces, the positions of facial “landmarks and upper torsos in images or video frames,” adding this deep neural network doesn’t “map the coordinates of the face” but determines “coordinates and size of a bounding box” so it can choose the ideal cropped portion of the full image.
“The location of a face within a photo says nothing about that face’s geometry, and plaintiffs have not identified any record evidence suggesting that this information would be sufficient to identify a person,” Rosenstengel wrote. “As a result, the iCloud Subclass, which claims only that face crops, user-inputted information, and other metadata was synced to iCloud, does not have a viable legal theory on which to proceed.”
She decertified that subclass, while noting its members still remain part of a Local Device Class, then explained her rejection of Apple’s argument as it pertains to faceprints.
“Because the app is merely comparing photographs, not creating, collecting or comparing face geometry, Apple argues, BIPA does not apply,” Rosenstengel wrote, stating the company was downplaying “the significance of its technology.”
Apple argued its two neural networks — deep and convolutional — truly only use “two-dimensional patterns of color and intensity” in pixels rather than a human face’s actual three-dimensional shape, adding the photos plotted into software vectors could never “be used to recreate a face.”
But the plaintiffs countered by claiming the convolutional network “mimics human vision by employing convolutional layers that detect and interpret geometric patterns in visual data,” Rosenstengel wrote, comparing the situation to Meta’s arguments in BIPA litigation in San Francisco federal court regarding Facebook’s suggested photo tags feature.
“Apple suggests that it merely ‘scans’ photographs, not faces, and compares the photographs to sort them into albums,” she wrote. “It also asserts that, under BIPA, ‘face geometry’ means a ‘three-dimensional representation of the facial structure.’ Like Facebook, Apple has construed the terms ‘scan’ and ‘face geometry’ too narrowly.”
The plaintiffs supplied expert reports about how 2D photos can have encoded information within pixels regarding 3D surfaces, including the opinion that certain deep learning models can indeed generate a useful 3D face representation and faceprints a convolutional network makes can be reverse engineered to reconstruct face images.
“Even Apple’s biometrics expert, Dr. Creed Jones, agreed that CNNs ‘operate directly’ on the pixels on an image, ‘enabling them to capture complex, non-linear relationships between images of faces,’ ” Rosenstengel wrote. “Moreover, Dr. Jones opines, ‘face recognition algorithms are designed to maximize the distinctions between human faces, to maximize the model’s ability to distinguish between people even if they appear superficially similar.’ To do this, the CNN ‘has as its input the entire face image, and each individual element in the embedding (faceprint) may be calculated from many different parts of the face image.’ In other words, Dr. Jones does not dispute that CNNs ‘scan’ the pixels of an image to create a faceprint that encapsulates the many different features of a face. Nor does he offer an alternative theory of how the CNN is able to distinguish between faces other than some reliance on geometry in its broadest sense.”
Because a reasonable jury could take that information — “along with the plethora of other exhibits including internal documents, patent applications, expert opinions, and testimony from Apple employees” — to reach the plaintiffs’ conclusion, Rosenstengel said, summary judgment for Apple is improper.
But it also is too early to hand the plaintiffs a win, she continued, because Apple has provided sufficient competing evidence.
“For example, Dr. Jones opined that faceprints do not represent measurements of a person’s face, are not based on facial landmarks and do not measure or capture the geometric representation of a face,” Rosenstengel wrote. “He also explains in his rebuttal report that the ‘shapes recognized by CNNs are more accurately described as local intensity patterns in the pixels of a photograph than geometric shapes’ and that the calculation performed by a convolutional layer is a ‘sum-of-products response to intensity patterns,’ not a geometric calculation.”
She also said a reasonable jury could find faceprints are capable of identifying a person and singled out the iCloud subclass where those prints are synced with other “significant identifying information.” But she again noted a jury could side with Apple, such as it if accepted the undisputed framing of a faceprint as “a 128-dimensional vector of decimal-point numbers, not an image of a face.”
Apple further agued it has no access to data stored on users’ devices and therefore doesn’t control any BIPA-protected information. Rosenstengel said that position falters in light of the July 2024 software update that began syncing device-based prints with iCloud servers, which might actually allow for plaintiffs to argue Apple does control locally-stored files.
“Apple concedes that faceprints are ‘transmitted and stored’ in the iCloud Photos library for users who meet a specific list of criteria,” Rosenstengel wrote. “But its insistence that it cannot unscramble and decipher the data is immaterial to its ‘control’ of the data. By dint of its authority over its servers where the data is maintained and its software, it undoubtedly could (among other things) re-write, make available to the public, remove encryption from or delete entirely, the faceprint data. Apple admitted as much at oral argument, when it conceded that it could decide one day to no longer protect the data with end-to-end encryption. And technological progress may make today’s encryption flimsy.”
Unlike willingly opting-in to facial or fingerprint recognition for device security purposes, she said, users of the software at question in this litigation “do not consent to the collection of their faceprints, they do not know the faceprints are being collected, there is no ability for users to turn off the feature, and there is no user knowledge that faceprints are saved on the device or in the cloud.”
Finally, she said BIPA doesn’t require a plaintiff to show a legal harm outside of proving a company possessed or controlled protected information without full compliance.
Named plaintiffs include Richard Robinson, of Troy; Yolanda Brown, of Godfrey; Jonathan LeBlond, of Milstadt; Patricia Orris, of Granite City; Angela Stevens, of West Frankfort; Jessica Jackson, of Metropolis; Melissa Oatman, of Collinsville; Justin Revelo, of Fairfield; and Taylor Vall, of Springfield.
Plaintiffs are represented by attorneys Andrew D. Schlichter, Jerome J. Schlichter, Troy A. Doles, Alexander L. Braitberg and Chen Kasher, of the firm of Schlichter Bogard, of St. Louis and Chicago; and Christian G. Montroy, of Montroy Law Offices, of Maryville.
Apple is represented by attorneys Purvi G. Patel, Emma Burgoon, Katie Viggiani and Tiffani B. Figueroa, of Morrison & Foerster, of Los Angeles and New York.
Sunny
