An Empirical Evaluation of GDPR Compliance Violations in Android mHealth Apps

Ming Fan, Le Yu, Sen Chen, Hao Zhou, Xiapu Luo, Shuyue Li, Yang Liu, Jun Liu, Ting Liu

TL;DR

This paper introduces HPDROID, an automated system to detect GDPR compliance violations in Android mHealth apps. By analyzing privacy policies and app code, they found that a significant portion of mHealth apps have incomplete privacy policies, inconsistent data collection practices, and insecure data transmission, posing serious privacy risks to users.

What Problem Does This Paper Address?

The General Data Protection Regulation (GDPR) aims to enhance privacy protection for EU citizens. However, GDPR provides general rules rather than specific implementation guidelines, creating a "semantic gap" between the regulation and actual app development. This gap can lead to compliance violations, exposing users to privacy threats. Mobile health (mHealth) apps are particularly concerning because they often collect sensitive Personal Health Information (PHI). Existing tools struggle to bridge this gap by automatically analyzing both privacy policies and app code to identify these violations.

Methods

The authors developed an automated system called HPDROID, comprising three main modules:

  • Incomplete Privacy Policy Detection: This module uses Natural Language Processing (NLP) to parse privacy policies and machine learning classifiers (Random Forest and Decision Tree) to identify six key notice categories required by GDPR (Data Collection, Data Usage, User Right, User Consent, Data Security, Contact Information).
  • Inconsistent Behavior Analysis: This module first identifies user-input PHI from the app's Graphical User Interface (GUI) using techniques like hint-based and label-based methods. Then, it employs data-flow analysis to track how this PHI is stored or transmitted. Finally, it compares the collected PHI (ACP) with the declared PHI in the privacy policy (DCP) to detect inconsistencies.
  • Insecure Data Transmission Identification: This module analyzes apps for the use of encryption and Secure Socket Layer (SSL)/Transport Layer Security (TLS) protocols. It checks for common encryption misuses (e.g., ECB mode, weak hashing algorithms, constant IVs/keys) and SSL vulnerabilities (e.g., inadequate certificate validation).

Datasets

  • Training Data: A manually labeled dataset of 1,284 sentences from 100 mHealth app privacy policies was created to train the notice classifiers. Each sentence was labeled with one or more of the six GDPR notice categories.
  • Evaluation Data: The study evaluated 796 real mHealth apps crawled from the Google Play Store. Apps with non-English privacy policies were excluded.

Key Insights and Ideas

  • The "Semantic Gap" is Real and Pervasive: The study highlights that the abstract nature of GDPR rules makes it difficult for developers to implement them correctly, leading to widespread violations.
  • mHealth Apps are Particularly Vulnerable: Due to the sensitive nature of the data they handle, mHealth apps are a critical area for privacy compliance, and this study reveals significant issues.
  • Privacy Policies are Often Incomplete: Developers frequently fail to include all necessary information in their privacy policies, leaving users uninformed about data practices.
  • Data Collection is Often Inconsistent: Apps collect more data than they disclose in their privacy policies, a direct violation of data minimization principles.
  • Data Transmission Security is Poorly Implemented: Even when apps attempt to secure data transmission, they often do so with critical misuses, leaving data vulnerable to breaches.
  • Automated Analysis is Crucial: Manual inspection is infeasible for the vast number of apps, making automated tools like HPDROID essential for identifying these complex violations.

Key Results

  • Incomplete Privacy Policies: 189 out of 796 mHealth apps (23.7%) provided incomplete privacy policies.
  • Inconsistent Data Collection: Of the 59 apps that collect user-input PHI, 46 (77.9%) exhibited at least one inconsistent data collection behavior (collecting more PHI than declared).
  • Insecure Data Transmission: Of the 59 apps that collect PHI, only 8 attempted to ensure transmission security. However, all 8 apps contained at least one encryption or SSL misuse.

Limitations and Future Work

  • Data-flow Analysis Limitations: The study acknowledges that data-flow analysis can have false negatives, meaning some PHI collection might be missed. Combining static analysis with dynamic analysis is suggested for future work to improve accuracy.
  • Self-Implemented Encryption: HPDROID's current encryption analysis relies on system-provided APIs. It cannot detect violations in self-implemented encryption functions. Future work could involve analyzing data entropy to identify potential encryption.
  • Ground Truth Dataset: While a triple-module redundancy approach was used for ground truth, inconsistencies among authors could lead to missing similar sentences in the training set, potentially affecting classifier performance.
  • Vague Policy Descriptions: The study found that vague descriptions like "other information" contribute to inconsistent data collection. Developers are urged to be more specific.
  • App User Consent: Many apps assume user consent by default, rather than actively acquiring it, which is a violation of GDPR.

The authors hope their tool will help developers identify and fix privacy issues before releasing apps and empower users to make more informed decisions.

Summary updated

BibTeX
@inproceedings{fan2020empirical,
  author = {Ming Fan and Le Yu and Sen Chen and Hao Zhou and Xiapu Luo and Shuyue Li and Yang Liu and Jun Liu and Ting Liu},
  title = {{An Empirical Evaluation of GDPR Compliance Violations in Android mHealth Apps}},
  booktitle = {ISSRE},
  year = {2020},
  doi = {10.1109/issre5003.2020.00032},
}

Loading related content...

Figure