Federated Learning of Cohorts (FLoC) is a type of web tracking. It groups people into "cohorts" based on their browsing history for the purpose of interest-based advertising. FLoC was being developed as a part of Google's Privacy Sandbox initiative, which includes several other advertising-related technologies with bird-themed names. Despite "federated learning" in the name, FLoC does not utilize any federated learning.
Google began testing the technology in Chrome 89 released in March 2021 as a replacement for third-party cookies. By April 2021, every major browser aside from Google Chrome that is based on Google's open-source Chromium platform had declined to implement FLoC. The technology was criticized on privacy grounds by groups including the Electronic Frontier Foundation and DuckDuckGo, and has been described as anti-competitive; it generated an antitrust response in multiple countries as well as questions about General Data Protection Regulation compliance. In July 2021, Google quietly suspended development of FLoC; Chrome 93, released on August 31, 2021, became the first version which disabled FLoC, but did not remove the internal programming.
On January 25, 2022, Google officially announced it had ended development of FLoC technologies and proposed the new Topics API to replace it. Media and competitors (e.g., Brave) criticized Topics API as a rebranding of FLoC with only minor changes and without addressing any main concerns.
Function
The Federated Learning of Cohorts algorithm analyzes users' online activity within the browser, and generates a "cohort ID" using the SimHash algorithm to group a given user with other users who access similar content. Each cohort contains several thousand users in order to make identifying individual users more difficult, and cohorts are updated weekly. Websites are then able to access the cohort ID using an API and determine what advertisements to serve. Google does not label cohorts based on interest beyond grouping users and assigning an ID, so advertisers need to determine the user types of each cohort on their own.
Opting out of cohort calculation
FLoC experiment was active only in Google Chrome browser and ran from Chrome 89 (inclusive) to Chrome 93 (not inclusive). Modern browsers do not support FLoC. While the experiment was active, users could opt out of FLoC experiment by disabling third-party cookies. Website administrators could opt out from cohort calculation via special HTTP headers. It can be accomplished with a new interest-cohort permissions policy or feature policy, the default behavior is to allow cohort calculation. To opt-out of all FLoC cohort calculations a website could send either of the following HTTP response headers: Permissions-Policy: browsing-topics=() or Feature-Policy: browsing-topics 'none' Google Chrome applies interest-cohort Feature Policy restrictions to Browsing Topics API as well.
Timeline
Initial prototype
On August 22, 2019, Google Chrome developers coined the term FLoC and first started discussing the upcoming replacement for cookies. In July 2020, the United Kingdom's Competition and Markets Authority found the FLoC proposal to be anti-competitive, since it would "place the browser in a vital gatekeeper position for the adtech ecosystem." Instead, the authority recommended adoption of a competing proposal called SPARROW, which maintains the same privacy-enhancing objectives but creates a different completely independent "Gatekeeper" which does not have any other role in the adtech ecosystem and does not have access to user-level information.
Testing
Google began testing FLoC in the Chrome 89 released in March 2021 as a replacement for third-party cookies, which Google plans to stop supporting in Chrome by mid-2023. (Initially Google announced plans to remove third-party cookies by late 2021, then postponed it to early 2022, and then to 2023 due to delay of FLoC technology.) The initial trial turned on FLoC for 0.5% of Chrome users across 10 countries: the United States, Australia, Brazil, Canada, India, Indonesia, Japan, Mexico, New Zealand and the Philippines. Users were automatically placed in the trial and were not notified, but could opt out by turning off third-party cookies. Furthermore, site administrators could disable FLoC and opt out from interest calculation via a Feature-Policy header. The initial trial did not include users in the United Kingdom or the European Economic Area due to concerns about legality under the area's privacy regulations.