Usability and Security Perceptions of Implicit Authentication: Convenient, Secure, Sometimes Annoying

Hassan Khan, Urs Hengartner, Daniel VogelSOUPS 2015 · OpenReview

TL;DR

This paper investigates the real-world usability and security perceptions of implicit authentication (IA) on smartphones. It reveals that while IA is perceived as convenient and secure by most users, false rejects can lead to annoyance, and a lack of transparency about its operation can raise security concerns. The findings offer crucial insights for designing more effective and user-friendly IA systems.

What Problem Does This Paper Address?

Many smartphone users forgo traditional explicit authentication (like PINs or patterns) due to inconvenience, leaving their devices vulnerable. Implicit authentication (IA) offers a potential solution by continuously and transparently recognizing users based on their behavioral biometrics (e.g., typing patterns, touch gestures). While IA is often touted as more usable, there's a lack of empirical evidence to support these claims and understand user perceptions of its security. This paper aims to bridge this gap by empirically evaluating IA's usability and security from the user's perspective.

Methods

The study employed a two-part design:

  1. Controlled Lab Experiment: Participants performed simulated smartphone tasks under IA and their preferred explicit authentication (EA) or no authentication. This allowed for controlled, quantitative measurement of performance metrics like task completion time and error rates.
  2. Field Study: Participants used a simulated IA system on their own smartphones for three days. This provided a more ecologically valid understanding of annoyance caused by IA's false rejects (FRs) and allowed for the exploration of different operating thresholds.

A pseudo-IA scheme was used in both parts to simulate realistic false reject rates and control experimental variables, particularly due to Android's limitations in accessing raw touch data without rooting. Participants provided feedback through surveys and semi-structured interviews.

Datasets

Not discussed in the paper. The study relied on simulated data and participant feedback rather than pre-existing datasets for training or evaluation.

Key Insights and Ideas

  • Convenience is King (and IA Delivers): A significant majority (91%) of participants found IA convenient, with many reporting it as more convenient than their existing EA methods. This directly addresses the primary reason users avoid traditional authentication.
  • Annoyance is Real, But Manageable: While IA is generally well-received, false rejects (where the system incorrectly flags a legitimate user as an imposter) were a source of annoyance for 35% of participants. This annoyance was often linked to the unpredictability of these interruptions.
  • Security Perceptions are Mixed but Generally Positive: Most users (81%) felt the level of protection offered by IA was satisfactory. However, concerns about false accepts (where an imposter is incorrectly authenticated) and detection delays were noted by a significant minority.
  • Transparency Matters: A key finding is that users felt less secure when IA was completely opaque (running as a background service without clear indicators). The lack of visible confirmation that IA was active raised concerns about whether it was actually protecting their device.
  • User Control is Desired: Participants expressed interest in customizing IA's operating threshold, allowing them to balance security and convenience based on their current needs and environment.

Key Results

  • Usability:
    • IA decreased overall task completion time by 7.1% for users who already used authentication (UseAuth).
  • IA did not increase task error rates.
  • IA was rated as significantly easier to use and more likely to be used frequently compared to EA.
  • 35% of participants found IA interruptions annoying, with this being more pronounced for users who already used EA.
  • Security:
    • 81% of participants were satisfied with the level of protection provided by IA.
  • Users who did not use any authentication (DontUseAuth) perceived IA as significantly more secure than EA.
  • Concerns were raised about false accepts (27%) and detection delays (22%).
  • 63% of participants were interested in adopting IA, either as a primary or secondary authentication method.

Limitations and Future Work

The authors acknowledge several limitations:

  • Pseudo-IA Scheme: The use of a simulated IA system, while necessary for control, meant participants might have experienced unexpected behaviors (e.g., FRs or TAs for similar inputs).
  • Field Study Context: For users already employing EA, IA was tested as a secondary layer, leading to potentially redundant authentications that could inflate annoyance.
  • In-Situ Feedback: The requirement for participants to provide feedback after each interruption might have increased annoyance.
  • Limited Comparison: The study did not directly compare IA with biometric-based EA schemes like fingerprint or facial recognition.

Future work suggested includes:

  • Exploring alternative methods to mitigate interruption annoyance, such as visual indicators or less intrusive authentication prompts.
  • Investigating the design of IA status indicators to improve transparency without creating new security vulnerabilities.
  • Developing user-adjustable operating thresholds with clear communication of the security-annoyance trade-off.
  • Conducting studies that directly compare IA with biometric-based EA systems.

Summary updated

BibTeX
@inproceedings{khan2015usability,
  author = {Hassan Khan and Urs Hengartner and Daniel Vogel},
  title = {{Usability and Security Perceptions of Implicit Authentication: Convenient, Secure, Sometimes Annoying}},
  booktitle = {SOUPS},
  year = {2015},
  url = {https://openreview.net/forum?id=eRC1IDLECi},
}
Figure