Aquilis: Using Contextual Integrity for Privacy Protection on Mobile Devices
summary
A privacy-preserving keyboard gives color-coded warnings using a probabilistic model based on pairwise mutual information and contextual integrity tuples to score instantaneous, longitudinal and cross-platform risk. It stores longitudinal and cross-platform knowledge graphs on-device (recent history only) and never blocks input. On 1,080 labeled texts Aquilis reports F1 = 0.7605; a 35-person study shows 74.3% agreement and users rate usefulness ~4.4/5 and subjective match 3.87/5. Heeding 50% of warnings cuts inappropriate sharing up to 30% while repeated exposure raises risk by ~340%.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 0:15Motivation
- 1:10Contributions
- 2:55Sensitivity Score
- 3:35Knowledge Graphs
- 4:13Accuracy
- 5:03Conclusion
related talks
41mSome Applications of Contextual Integrity
01h 04mAcademic sessions privacy and surveillance
20mEnhancing Mobile Content Privacy with Proxemics Aware Notifications and Protection
06mEvaluating and Redefining Smartphone Permissions with Contextualized Justifications for
53mDesigning for Privacy in Mobile and Web Apps
17mMeasuring Privacy Risks in Mobile Apps at Scale
22mEnhancing Privacy through an Interactive On-demand Incremental Information
10mCommunicating Differential Privacy Guarantees to Data Subjects