Facial recognition technology (“FRT”) has attracted a fair amount of attention over the years, including in the EU (e.g., see our posts on the European Parliament vote and CNIL guidance), the UK (e.g., ICO opinion and High Court decision) and the U.S. (e.g., Washington state and NTIA guidelines). This post summarizes two recent developments in this space: (i) the UK Information Commissioner’s Office (“ICO”)’s announcement of a £7.5-million fine and enforcement notice against Clearview AI (“Clearview”), and (ii) the EDPB’s release of draft guidelines on the use of FRT in law enforcement.
I. ICO Fines Clearview AI £7.5m
In the past year, Clearview has been subject to investigations into its data processing activities by the French and Italian authorities, and a joint investigation by the ICO and the Australian Information Commissioner. All four regulators held that Clearview’s processing of biometric data scraped from over 20 billion facial images from across the internet, including from social media sites, breached data protection laws.
On 26 May 2022, the ICO released its monetary penalty notice and enforcement notice against Clearview. The ICO concluded that Clearview’s activities infringed a number of the GDPR and UK GDPR’s provisions, including:
- Failing to process data in a way that is fair and transparent under Article 5(1)(a) GDPR. The ICO concluded that people were not made aware or would not reasonably expect their images to be scraped, added to a worldwide database, and made available to a wide range of customers for the purpose of matching images on the company’s database.
- Failing to process data in a way that is lawful under the GDPR. The ICO ruled that Clearview’s processing did not meet any of the conditions for lawful processing set out in Article 6, nor, for biometric data, in Article 9(2) GDPR.
- Failing to have a data retention policy and thus being unable to ensure that personal data are not retained for longer than necessary under Article 5(1)(e) GDPR. There was no indication as to when (or whether) any images are ever removed from Clearview’s database.
- Failing to provide data subjects with the necessary information under Article 14 GDPR. According to the ICO’s investigation, the only way in which data subjects could obtain that information was by contacting Clearview and directly requesting it.
- Impeding the exercise of data subject rights under Articles 15, 16, 17, 21 and 22 GDPR. In order to exercise these rights, data subjects needed to provide Clearview with additional personal data, by providing a photograph of themselves that can be matched against the Clearview Database.
- Failing to conduct a Data Protection Impact Assessment (“DPIA”) under Article 35 GDPR. The ICO found that Clearview failed at any time to conduct a DPIA in respect of its processing of the personal data of UK residents.
The ICO decided to fine Clearview £7.5 million, in contrast to the £17 million fine initially suggested. The monetary penalty notice sets out the ICO’s reasoning behind the fine. As Clearview did not provide any figures for its income or turnover, the ICO was unable to calculate Clearview’s financial gain from the activities in question, which would ordinarily serve as a step in calculating the fine amount. The ICO had regard to the range of penalties available to it, and set an initial amount at just below the mid-point of this range, amounting to £7,552,800. The ICO then considered the other statutory factors, and did not consider that any of these justified either an increase or a reduction from the initial starting point. This included consideration of certain representations made by Clearview that it had acted on requests from UK data subjects to exclude their images from future searches.
Although Clearview no longer offers its services to UK organisations, due to the perceived risk of data concerning UK residents being used in the company’s offerings elsewhere, the ICO ordered Clearview to take the following steps in addition to paying the £7.5 million fine:
- Deleting the personal data of UK residents from its systems, within six months of the expiry of the appeal period.
- Refraining from any further processing of the personal data of data subjects resident in the UK, within three months following the date of the expiry of the appeal period.
- Refraining from offering any service provided by way of its database to any UK customer.
- Refraining from doing anything in the future that would fall under the above points without first drafting a DPIA, and providing this to the ICO.
The ICO imposed a similar deletion order on HMRC in 2019, in which it gave the tax authority 28 days to delete all biometric voice data for which it did not have explicit consent to process.
II. EDPB Publishes New Draft Guidelines on the Use of FRT in Law Enforcement
On 12 May 2022, the EDPB adopted draft guidelines providing guidance to law makers and law enforcement authorities (“LEAs”) on implementing and using FRT systems. The guidelines provide that FRT should only be used in compliance with the Law Enforcement Directive (“LED”) and only in a necessary and proportionate manner, as set out in the Charter of Fundamental Rights. The EDPB’s draft guidelines are currently open for public consultation until 27 June. If adopted, they will impact the requests for data, software, and other technology that the EU and LEAs can make of private companies.
The EDPB makes several references to processing of personal data in a law enforcement context that relies on databases, similar to Clearview’s, populated by “scraping” photographs accessible online on a mass scale, including calling for a ban on LEAs’ use of such databases. The EDPB notes that when assessing whether processing relates to data which are “manifestly made public by the data subject” (a lawful ground for processing biometric data under Article 10 LED), the fact that a photograph is “manifestly made public” does not mean that the related biometric data which can be retrieved from the photograph using FRT tools has also been “manifestly made public”. For biometric data to be seen as “manifestly made public,” the data subject must have deliberately made their biometric data freely accessible and public through an open source. Further, the EDPB notes that default settings of a service (e.g., if data is made public by default on a social networking platform) should not be construed as data “manifestly made public.”
The guidelines repeat the EDPB’s prior call for a ban on the use of FRT in certain cases, specifically:
- Remote biometric identification of individuals in publicly accessible spaces;
- FRT categorizing individuals based on their biometrics into clusters according to ethnicity, gender, as well as political or sexual orientation or other grounds for discrimination;
- FRT to infer emotions of a natural person; and
- As described above, processing of personal data in a law enforcement context that relies on a database populated by the collection of personal data on a mass-scale and in an indiscriminate way.
The EU is also currently debating whether to prohibit certain forms of “real time” remote biometric identification systems in the context of its proposal for a Regulation laying down harmonized rules on artificial intelligence (the “EU AI Act”) (see our blog here for further details). The EDPB and the European Data Protection Supervisor published a Joint Opinion on the EU AI Act last year.