For over a century, the right of publicity (ROP) has protected individuals from unwanted commercial exploitation of their identities. Originating around the turn of the twentieth century in response to the newest image-appropriation technologies of the time, including portrait photography, mass-production packaging, and a ubiquitous printing press, the ROP has continued to evolve to cover each new wave of technologies enabling companies to exploit peoples' identities as part of their business models. These evolutionary moments for the ROP made sense in light of these new technologies and their capacity to enable mass appropriation of images and identities. For example, in the 1890s, advances in printing and photochemical technologies led publishers to inundate modern society with mass-circulation magazines and newspapers, increasingly adorned with images of people in advertisements and photographs accompanying stories. Barbas,supra,at 10,48. As the demand for more and more images grew, supply remained stagnant; popular mass photography and a commercial modeling industry were still decades away. Id. at 49. At the same time, ordinary people were flocking to portrait photography studios for personal memories and keepsakes. As photographers soon realized, their archives of portrait negatives and prints had subsequent commercial value. Soon, a tremendous black market emerged for these images, with photographers regularly supplying images without the consent of their subjects, most of whom were ordinary people "whose images were fungible and ubiquitous and who would be at 56. The present case against Clearview involves the same type of appropriation, taken from internet websites instead of portrait photographers' studios.
The elements of California's common law ROP track this approach, requiring a showing that: (1) a defendant used the person's image or identity;
(2) the appropriated image or identity was used to the violator's advantage, "commercially or otherwise"; (3) a lack of consent; and (4) Feb. 24, 2010) (quoting Eastwood v. Sup. Ct., 198 Cal. Rptr. 342, 347 (Ct. App. 1983)). As noted below, Clearview's development and deployment of its facial recognition (FR) app satisfy all four elements.
According to its own description, Clearview directly uses individuals' images and identities to build and operate its facial recognition (FR) app.
To determine whether an alleged violator "uses" an image or identity under the ROP, a court simply looks to whether the defendant was responsible for the alleged use. See Fleet v. CBS, Inc., 50 Cal. App. 4th 1911, 1918(1996) (citing Restatement (Third) of Unfair Competition § 46 (Am. L. Inst. 1995)). "Use" is often obvious where a defendant took the direct action that violated the ROP.
In California, the ROP "does not require that appropriations of identity be accomplished through particular means to be actionable." White v. Samsung Elecs. Am., Inc., 971 F.2d 1395, 1398(9th Cir. 1997); see also Eastwood v. Superior Court, 198 Cal. Rptr. 342 (Ct. App. 1983); No Doubt v. Activision Publ'g, Inc., 122 Cal. Rptr. 3d 397 (Ct. App. 2011). Courts apply the ROP to appropriations based on characteristics that have some clearly recognizable association with a particular person, even in the absence of their name or image. See White, id.
Part of California's broad ROP protections is the idea that an identity can be appropriated no matter how it is used. In several cases, courts applied the ROP to appropriations of identity where the identity itself was or was a part of the product, rather than in an advertisement for a separate product. See, e.g., Comedy III Prods. v. Saderup, 25 Cal. 4th 387, 395 (2001) ii.
Clearview uses individuals' images to create facial vectors that train its FR algorithm to accurately identify individuals and that enable the algorithm to identify individuals from a probe image.
In the same way that nineteenth-century mass technologies created a new market of appropriated images for human audiences, machine learning technologies like FR have created a new market of appropriated images for both machine and human audiences. Clearview uses the scraped images in its database to train its FR algorithm with likenesses and facial vectors drawn from those images, increasing its accuracy and commercial viability. Without prior images that have been labeled with unique identities, Clearview's app would have no way to compare and recognize novel inputs.
Clearview's FR system also uses the millions of identities contained in the database each time it peruses through various facial vector neighborhoods, searching for a specific face. Even if the app only outputs a single positive identity match in response to a probe image, Clearview uses every individual image and identity each time its algorithm crunches a probe image into a facial vector and compares it to the sea of facial vectors in the database.
Without denying these uses, Clearview argues that any alleged ROP violations occur in a "black box," shielding its appropriation of images and identities from human observation. See Def.'s MPA in Supp. of Mot. to Strike at 12; see also Mulcaire Decl. ¶ ¶ 30-42. But the lack of a "human in the loop" does not negate Clearview's use of images and identities to perfect its FR algorithm. Today's emergent mass technologies increasingly rely on machine learning, and FR is a primary example of how the decades-old field of computer vision has advanced to the point where computers-not humans-are the primary audiences for our images. While most people assume that humans look at images, "and that the relationship between human viewers and images is the most important moment to analyze," computer vision enables an algorithm to "see" (Dec. 8, 2016), https://thenewinquiry.com/invisible-images-your-pictures-are-looking-at-you/.
Today, the "majority of images are now made by machines for other machines [to see]." A primary example of this shift is "the trillions of images that humans share on digital platforms."
When people upload their images online, they "feed[] an array of immensely powerful artificial intelligence systems information about how to identify people." Id. Thus, whether the audience is ultimately a machine or a human or both makes no difference for this Court's ROP analysis: use for commercial gain is the key test of Clearview's conduct.
Clearview uses individuals' images to display a facial recognition match to the end user of its app.
Finally, and most obviously, Clearview uses individuals' identities when it produces a match. When the app identifies an individual, it provides a gallery of images that align with the facial vectors in the input photo. The result page also includes links to where those photos appeared originally, meaning the sites from which Clearview scraped the image. See Hill, End Privacy, supra. The result is the individual's identity-that is the point of the product. Its use of identity in this way is far from incidental. Cf. Def.'s Demurrer at 11 n.4. It is the intended outcome of Clearview's FR app and the app's main selling point.
Clearview appropriates individuals' identities by capturing incorporating, and commercially exploiting their unique facial attributes and identities from their facial images.
Clearview not only appropriated the images of peoples' faces which comprise its massive face database, but it also appropriated their identities by constructing facial vectors that can uniquely identify a particular individual. The ROP allows broad liability for the appropriation of any characteristic that has a clearly recognizable association with an individual. See, e.g., Zacchini v. Scripps-Howard Broadcasting Co., 433 U.S. 562 (1977) Clearview's FR algorithm constructs recognizable associations based on the facial vector of everyone in its vast system-the facial vectors must map onto an individual's unique identity accurately for the app to have any value. More faces mean more accuracy, more accuracy means more value provided to customers, more value entices more customers, and more customers mean more profit overall. Clearview's entire business strategy is to profit off its mass appropriation of identities, and those identities are the lifeblood of its commercial success. identify an individual from a new image. That Clearview may have designed its app to appear like a search engine to its customers does not magically convert its powerful FR technology into a search engine. Clearview openly promotes its FR product as a FR product-not a search engine that thrives on others' content, but a massive, closed universe of identities that only Clearview customers can access by paying. And Clearview admits that it has purposefully scrapped and ingested billions of individuals' images and identities into its database, used those images and identities to train its FR algorithm, to perform "matches," and as outputs to its customers. Such uses are hardly the work of third parties.
Placing one's images on the internet is not consent for Clearview to commercially appropriate such images or identities into its FR app.
Clearview does not attempt to argue it had consent from the millions of individuals whose images and identities it exploits in its FR app, and it would be extremely difficult to do so.
Consent must be knowing and use-specific. See, e.g., Cohen v. Facebook, Inc., 798 F. Supp. 2d 1090, 1095(N.D. Cal. 2011) (holding users did not consent to commercial use of their identities by using a Facebook service whose terms of use were too ambiguous to find consent); see also Pratt v. Everalbum,Inc.,283 F. Supp. 3d By upholding a viable ROP claim, this Court protects the same autonomy interests the right has historically covered: control over the use of one's identity by another for profit.
Clearview's app exacts a mass harm on an immense population of people affected by its harvesting of their images without consent. Clearview provides a quintessential example of the "theft of good will" that attends the use of something for nothing.
The anti-SLAPP statute applies only to legal claims "that arise[] from protected speech . . . and lack[] even minimal merit." Navellier v. Sletten, 29 Cal. 4th 82, 89 (2002) (emphasis in original). Courts must avoid the "fallacy that the anti-SLAPP statute allows a defendant to escape the consequences of wrongful conduct by asserting a spurious First Amendment defense." Id. at 93. As discussed above, this case raises an ROP claim with far more than minimal merit. Moreover, Clearview's portrayal of its proprietary FR product as protected speech fails as well. Clearview cannot hide behind its customers' downstream choices of how to use its app to excuse the misappropriations of image and identity that occur throughout the app's operation.
Clearview argues that its FR app is speech "identifying potential criminals" that is "squarely in the public interest." Def.'s Mem. in Supp. of Special Mot. at 8. In support of this, Case No.: RG21096898 AMICI CURIAE BRIEF OF SCIENCE, LEGAL, AND TECHNOLOGY SCHOLARS however, Clearview only cites a few examples of ways that a few of its customers choose to use its app. See id. at 8 (describing uses of app to identify perpetrators of the January 6th U.S.
Capitol attack, child sex traffickers and terrorists, and Russian occupiers of Ukraine). But how its customers choose to use its FR app cannot overcome Clearview's misappropriation of images and identities in its creation, design, and operation. For one thing, almost all commercial FR companies, like Clearview, train their FR algorithms completely under the veil of corporate secrecy and as far away from public discussion as possible. The only part of the process that is shared are the outputs of matched faces to customers; those matches are the commercial products that customers purchase in the first place-such "speech" is worlds away from expressive activities of protest or petition traditionally shielded under § 425.16. Clearview's argument allows potential public-interest uses to swallow its overall premise, which is approptiating images and identities belonging to millions of people who never consented to that use. Not even news publishers, who traditionally enjoy broad speech protections, have license "to invade the rights and liberties of others." Cohen v. Cowles Media Co., 501 U.S. 663, 670 (1991).
Further, Clearview does not provide even its law enforcement customers with any services during the early stages of the FR process. Clearview cherry-picks specific instances of police uses of its app to identify individuals involved in certain investigations. But even if those instances were protected by statute, the development and production of the app itself remains unprotected. When Clearview initially scraped billions of facial images from the internet, and when Clearview uses these images to construct facial vectors and train its algorithm on those constructions, it runs afoul of the ROP. The company's subsequent interactions with law enforcement customers, even if they are somehow public-interest oriented, cannot excuse these violations of law that predicated the app's success. Even if some downstream uses of Clearview's app are protected speech, Clearview's prior conduct cannot be so neatly excused.
At most, Clearview engages in conduct that "may conceivably have indirect consequences for an issue of public concern," depending on who uses its product and how. Rand Res., LLC v.
City of Carson, 6 Cal. 5th 610, 625 (2019). And yet, Clearview's speech is ostensibly performed