Conflicting Privacy Preference Signals in the Wild

Maximilian Hils, Daniel W. Woods, Rainer BöhmearXiv 2021 · OpenReview

TL;DR

This paper investigates how users express privacy preferences online, finding that the coexistence of multiple signals (like Global Privacy Control and cookie consent banners) leads to significant ambiguity. They provide the first empirical evidence that users send conflicting signals and that many users block cookie consent dialogs, especially those using privacy-enhancing browser features.

What Problem Does This Paper Address?

Modern privacy laws like GDPR and CCPA aim to give users control over their personal data. To facilitate this, web standards have emerged for users to communicate their preferences, such as the Global Privacy Control (GPC) and the Transparency & Consent Framework (TCF). However, the problem is that users can send multiple signals simultaneously, creating ambiguity. This ambiguity arises because different signals might convey conflicting information (e.g., a global "do not track" signal alongside accepting tracking in a website's cookie banner). This paper tackles the empirical question of how this ambiguity manifests in the wild and how it impacts privacy outcomes, especially in light of legal requirements for "unambiguous" consent.

Methods

The researchers embedded a logging script on three websites and collected data on user interactions with cookie consent dialogs (specifically, Quantcast's TCF implementation). They simultaneously logged whether users sent GPC or Do Not Track (DNT) signals via their browser. Their methodology involved:

  • Measuring Dialog Loading: Determining if the TCF consent dialog could be loaded, suspecting that blocking mechanisms (like ad-blockers) are a primary reason for failure.
  • Observing User Decisions: Recording whether users accepted or rejected tracking when presented with a TCF dialog.
  • Correlating Browser Signals with Dialog Behavior: Analyzing how the presence of GPC/DNT signals influenced users' decisions within the TCF dialog and whether the dialog itself was blocked.
  • Data Collection Pipeline: Utilizing browser page visibility API and the Beacon API to ensure complete event recording, even when pages are closed.

Datasets

The study collected data from impressions on three websites: mitmproxy.org (an open-source tool for developers), the authors' research group website, and a Capture The Flag contest website. This resulted in 16,761 impressions from 8,033 unique IPv4 addresses. The data was collected over a short period. No new benchmark datasets were introduced; rather, real-world user interactions were observed.

Key Insights and Ideas

  • Blocking is a Major Factor: A significant portion of users, especially those using privacy-enhancing browser features like GPC or DNT, actively block cookie consent dialogs. This "technical response" is a novel finding and represents a significant barrier to collecting explicit consent.
  • GPC Signals Genuine Preference: Users who enable GPC are significantly more likely to reject tracking in TCF dialogs, suggesting that the GPC signal is a reliable indicator of a user's privacy preferences.
  • Conflicting Signals are Common: Despite GPC users showing a stronger tendency to reject tracking, a substantial majority still accept tracking in TCF dialogs, highlighting the reality of conflicting signals.
  • DNT vs. GPC: While both DNT and GPC are single-bit signals, GPC appears to be a stronger indicator of privacy preference, with users being more likely to reject tracking. The paper notes that DNT and GPC signals are not statistically distinguishable in their effect on TCF choices in their sample.

Key Results

  • Dialog Blocking: 27% of users blocked the TCF consent dialog by default. This number increased to 50% for users with DNT enabled and 73% for users with GPC enabled.
  • GPC and Rejection: Users with GPC enabled were 1.9 to 2.0 times more likely to click "I do not accept" in a TCF dialog compared to users without GPC.
  • Acceptance Despite GPC: Despite the higher rejection rate, 73% of users with GPC enabled still clicked "I Accept" in the TCF dialog, demonstrating conflicting signals.
  • Prevalence of Signals: Approximately 3.5k impressions sent a DNT signal, and 550 sent a GPC signal. Most GPC signals were accompanied by a DNT signal.
  • Ambiguous Conflicts: Conflicts (accepting TCF while sending an opt-out signal) occurred in 5% of all impressions, and 77%/73% of DNT/GPC-enabled users, respectively.

Limitations and Future Work

  • Limited Website Scope: The study was conducted on only three websites, which likely over-represent technically literate and privacy-aware users. Future work could expand to a broader range of websites.
  • Browser Specificity: While the study observed GPC signals from various browsers, the exact origin of all Chromium-based GPC signals (e.g., Brave vs. other Chrome variants) could be further investigated.
  • Legal Interpretation: The paper defers the complex legal interpretation of these conflicting signals to legal analysis, acknowledging that different jurisdictions may resolve ambiguities differently.
  • Future Directions: The authors suggest further research into permanently storing privacy preferences in the browser, potentially with more than a single bit of information to accommodate diverse legal requirements. They also highlight the potential for an "arms race" as websites develop countermeasures against blocking mechanisms.

Summary updated

Abstract

Privacy preference signals allow users to express preferences over how their personal data is processed. These signals become important in determining privacy outcomes when they reference an enforceable legal basis, as is the case with recent signals such as the Global Privacy Control and the Transparency & Consent Framework. However, the coexistence of multiple privacy preference signals creates ambiguity as users may transmit more than one signal. This paper collects evidence about ambiguity flowing from the aforementioned two signals and the historic Do Not Track signal. We provide the first empirical evidence that ambiguous signals are sent by web users in the wild. We also show that preferences stored in the browser are reliable predictors of privacy preferences expressed in web dialogs. Finally, we provide the first evidence that popular cookie dialogs are blocked by the majority of users who adopted the Do Not Track and Global Privacy Control standards. These empirical results inform forthcoming legal debates about how to interpret privacy preference signals.
BibTeX
@misc{hils2021conflicting,
  author = {Maximilian Hils and Daniel W. Woods and Rainer Böhme},
  title = {{Conflicting Privacy Preference Signals in the Wild}},
  year = {2021},
  eprint = {2109.14286},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2109.14286},
}
Figure