Google Built a Surveillance System Into Its Chrome Browser to “Improve Privacy”
Google has announced that its Privacy Sandbox for the Web is now generally available to anyone using Chrome browser. It’s the latest iteration of Google’s efforts to improve online privacy without harming its digital advertising ecosystem, which currently provides around $30 billion of its revenue each year.
This blog first wrote about Google’s Privacy Sandbox in 2019. In 2021, this initiative became the Federated Learning of Cohorts (FLoC) technology. Following criticism of the latter’s approach from various quarters, FLoC has now been replaced by Privacy Sandbox once again.
One of the key aims of Google’s Privacy Sandbox is to move away from third-party cookies. These are small pieces of data that advertisers place in your system when you visit their sites, usually via a browser. Cookies allow you to be tracked as you move between sites, enabling the creation of huge, privacy-damaging databases of highly personal information. These databases are then used for micro-targeted advertising, typically based on real-time bidding that results in your data being sent to hundreds of companies with no controls or protection. Google’s plan to switch off support for third-party cookies from next year is theoretically good for privacy, but the big question now is whether or not Privacy Sandbox is significantly better.
There are three elements to the new approach:
- Protected Audience
Protected Audience is an updated version of the third-party cookies that currently enable companies to show you ads based on the sites you visit. The biggest problem with today’s approach is that huge stores of information are created on external servers, which also manage which ads you’re shown. Google’s innovation is to bring both the database of visited sites and the ad management function into its Chrome browser. This way, there will be no external database of personal information outside the user’s control, and the user can determine what information is passed to advertisers, instead of sending large datasets indiscriminately across the internet. However, it does mean that the Chrome browser is constantly monitoring what a user does online.
Topics is another way that advertisers can send relevant ads without being able to gather detailed knowledge about which sites you’ve visited. Each site is labelled with a standard set of high-level topics – things like gardening, wine, sport, etc. When a user moves around the web, Chrome collects the most frequent topics associated with the sites. These topics are then shared in a controlled way to help advertisers display relevant advertising. The topics that are sent can be viewed and deleted by the user if they wish. Once again, this means that Chrome is monitoring all sites visited.
Attribution is a way for advertisers to find out whether their ads resulted in purchases or other actions, but without providing personal details showing who bought what. Methods adopted to shield user information include encryption, time delays, data aggregation, and randomization. In principle, these sound like sensible techniques. Importantly, users can opt out of using them completely if they wish.
Not everyone is convinced by Google’s new approach. Apple doesn’t think Topics is a good idea for a variety of reasons. For example, the company doesn’t support creating cross-site data about users and then sharing that with advertisers, even in limited quantities. Apple says that “Users must have agency over expressing their personal interests to websites and third parties.” It also stated that “By default exposing cross-site data to facilitate personalized ad targeting would make the web less safe to roam. Users would have to always think twice about which sites they visit and how that can be used to manipulate or target them.” Apple also makes a very good point about how Topics, although apparently general enough, can provide deep insights in a person’s life:
topics connected to the user’s browsing will evolve over time, allowing continuous enrichment of the user profile as an ongoing privacy exposure. An example: The user was interested in honeymoons, then baby clothing, then lawyers.
Apple notes that it will be hard to come up with a set of Topics that is truly global and doesn’t just reflect an affluent Western lifestyle. Similarly, the definition of “sensitive” topics will vary across “cultures, religions, ages, communities and individuals,” making it hard to protect people in all situations. The increasing capabilities of AI systems mean that even seemingly vague topics may turn out to contain hidden patterns, and their discovery could harm privacy.
Mozilla is similarly concerned about Topics, writing: “Fundamentally, we just can’t see a way to make this work from a privacy standpoint.” Mozilla has published a detailed, and highly technical analysis of the methods Privacy Sandbox uses to obscure personal details, and points out a number of situations where it will be easy to extract personal information.
As a result, neither Apple nor Mozilla intend to support Google’s new approach in their browsers. Aside from these technical concerns, there are also antitrust issues. That’s something that the Movement for an Open Web is worried about, and it reflects the views of many in the adtech industry:
We estimate that our supporters work with more than 21,000 advertisers and nearly 6 million individual websites in more than 65 countries, jointly serving over 320 billion advertising impressions each year, and having over 63 million subscribers. The companies have a combined workforce of over 40,000 employees and combined revenues in excess of $40bn.
Many of these companies are worried by what they perceive as Google’s attempt to create a new adtech system which it will dominate even more completely than the current one.
The Privacy Sandbox does offer improvements over the current system, which is fundamentally incompatible with privacy, as numerous PIA blog posts have detailed. But it is not much better, involves constant in-browser surveillance, and leaves Google firmly in control. However, it does contain one element that points the way to a better solution.
The Topics approach categorizes sites based on the material they offer. Ads are then displayed according to some of the Topics that a visitor to that site has viewed. The same approach of categorizing sites could be used directly to choose relevant ads. The advent of AI systems could make this very easy to implement, as it means machine learning systems are now very good at ingesting huge quantities of material, then summarizing and categorizing it to produce detailed topics that could be used by advertisers.
This context-based approach to advertising is superior to Google’s Privacy Sandbox because it requires zero information from the user, and no built-in browser surveillance as Sandbox does. It simply shows ads that are relevant based on the material the user is viewing. That’s great for the user, and great for advertisers, since they can be sure that their ads are relevant. But it’s bad for Google, which is why the company will keep on coming up with clever ideas like the Privacy Sandbox that offer only incremental improvements, not thorough privacy protection for everyone online.
Featured image by Google.