TL;DR
This paper introduces Mimesis Aegis (M-Aegis), a novel system designed to protect user data privacy on public cloud services without compromising the user experience. By introducing a "Layer 7.5" between applications and users, M-Aegis provides end-to-end encryption and isolation, ensuring data confidentiality while maintaining the look and feel of existing apps.
What Problem Does This Paper Address?
In today's digital landscape, users increasingly rely on public cloud services (PCS) like Gmail, Facebook, and WhatsApp for daily communication and data storage. While convenient, these services present significant privacy risks. Users are implicitly forced to trust PCS providers with their sensitive data, a trust that is often misplaced due to legal obligations to share data with surveillance agencies, the PCS's own data mining business models, operator errors, and the potential for data server breaches. Existing solutions for end-to-end encryption (E2EE) often fall short on mobile platforms due to usability issues, the need for custom applications, or the complexity of integrating with diverse mobile apps. This paper addresses the challenge of providing robust data privacy on public clouds for mobile users without sacrificing the seamless user experience they've come to expect.
Methods
Mimesis Aegis (M-Aegis) introduces a conceptual "Layer 7.5" (L-7.5) that acts as an intermediary between the application layer (OSI Layer 7) and the user (Layer 8). This layer is implemented as a transparent, always-on-top window within an isolated process.
Key techniques include:
- User Interface Automation (UIA): M-Aegis leverages the Android Accessibility Service and its UIA framework to inspect and interact with the GUI of target applications. This allows M-Aegis to understand the app's layout, identify UI components (like text fields and buttons), and programmatically control them.
- Mimic GUIs: To preserve user experience, M-Aegis renders "mimic" GUIs that overlay the original application's interface. These mimic GUIs are visually distinct (e.g., different colors) to indicate that M-Aegis is active. User input is directed to these mimic GUIs, where it can be processed (e.g., encrypted) before being passed to the underlying application.
- Per-TCA Logic: For each target application (TCA), M-Aegis employs specific logic modules. These modules determine how to handle user input (e.g., encrypting sensitive data) and how to present encrypted output back to the user.
- Searchable Encryption (EDESE): To address the crucial functionality of searching within encrypted data without server-side modifications, M-Aegis incorporates a novel searchable encryption scheme called EDESE. This scheme uses Bloom filters to enable efficient, albeit susceptible to false positives, keyword searches on encrypted content.
Datasets
- Enron Email Dataset: Used for evaluating the performance of composing and sending encrypted emails. The dataset was sampled to represent typical email lengths and uniqueness of words.
- Not discussed in the paper: No new benchmarks were introduced.
Key Insights and Ideas
- Layer 7.5 as a Novel Abstraction: The introduction of L-7.5 is a key conceptual contribution, providing a flexible layer for privacy interventions without altering the underlying OS or applications.
- Mimicry for User Experience: The core idea of "mimicking" the application's UI is crucial for user adoption. By making the privacy solution visually and functionally transparent, users are less likely to resist its use.
- UIA as a Robust Interface: Leveraging Android's UIA framework proves to be a powerful and resilient method for interacting with diverse applications, even surviving app updates that change UI elements' appearance but not their underlying resource IDs.
- Searchable Encryption without Server Cooperation: The EDESE scheme is a significant contribution, enabling a critical feature (search) for encrypted data without requiring any changes to the cloud service provider's infrastructure, aligning with the untrusted PCS model.
Key Results
- Negligible Latency: M-Aegis imposes minimal overhead on user experience. Encryption/decryption latency is imperceptible, and search operations are nearly instantaneous. Previewing encrypted emails adds an average of 76 ms, well within the user-noticeable threshold.
- Composing/Sending Email: Encrypting an email and building the search index takes approximately 205 ms for a typical email, including network latency.
- User Acceptability: A user study with 15 participants indicated that users found no major differences in their overall experience using Gmail with M-Aegis. All participants stated they would use M-Aegis to protect their data privacy after understanding its purpose, suggesting high potential for adoption.
- Correctness: Manual verification confirmed that M-Aegis correctly handles encryption, decryption, and searchable encryption, with no false negatives in search results.
Limitations and Future Work
- Social Engineering Attacks: M-Aegis does not protect against social engineering attacks (e.g., phishing) that might trick users into disabling the system or entering sensitive information into unprotected UIs. The system offers visual cues (different UI colors) as a best-effort mitigation.
- Data Format Limitations: Currently, M-Aegis primarily supports text-based communication. Handling other data types like images, audio, or video is a future direction, requiring the development of appropriate transformation functions (encryption and encoding) that can handle compression and other media-specific processes.
- Typographical Errors in Search: The current searchable encryption scheme does not tolerate typographical errors in search queries. The authors note that mobile keyboards' spell-check features mitigate this in practice.
- Future Work: The authors suggest extending M-Aegis to support more cloud services and media types. They also plan to continue user studies to further validate acceptability and improve the prototype based on feedback. The potential for integrating with more robust security mechanisms like PGP key management is also mentioned.
Summary updated