Evidence · DiggingBeagle record
MobileInject: Cross-modal consistency injection in Android AI agents
MobileInject tests coordinated malicious screen pixels and accessibility metadata in Android-agent emulator tasks. The source reports 108 constructed risk instances and action-level evaluation. It does not document attacks against users of real commercial mobile applications.
- Source role
- primary disclosure
Evidence record
MobileInject tests coordinated malicious screen pixels and accessibility metadata in Android-agent emulator tasks. The source reports 108 constructed risk instances and action-level evaluation. It does not document attacks against users of real commercial mobile applications.
Claim-level citations (2)
- supportsMobileInject shows how matching screenshots and accessibility text can mislead mobile agents: MobileInject implements 108 Android-emulator risk instances across 11 application scenarios and nine risk categories.
Abstract; benchmark construction
- supportsMobileInject shows how matching screenshots and accessibility text can mislead mobile agents: The benchmark tests attacks that place semantically consistent malicious goals into visual and textual observations, which can share one attacker-controlled origin.
Cross-Modal Consistency Injection methods