Practical Data Access Minimization in Trigger-Action Platforms

Paper in proceeding, 2022

Trigger-Action Platforms (TAPs) connect disparate online services and enable users to create automation rules in diverse domains such as smart homes and business productivity. Unfortunately, the current design of TAPs is flawed from a privacy perspective, allowing unfettered access to sensitive user data. We point out that it suffers from two types of overprivilege: (1) attribute-level, where it has access to more data attributes than it needs for running user-created rules; and (2) token-level, where it has access to more APIs than it needs. To mitigate overprivilege and subsequent privacy concerns we design and implement minTAP, a practical approach to data access minimization in TAPs. Our key insight is that the semantics of a user-created automation rule implicitly specifies the minimal amount of data it needs. This allows minTAP to leverage language-based data minimization to apply the principle of least-privilege by releasing only the necessary attributes of user data to TAPs and fending off unrelated API access. Using real user-created rules on the popular IFTTT TAP, we demonstrate that minTAP sanitizes a median of 4 sensitive data attributes per rule, with modest performance overhead and without modifying IFTTT.