Adaptive Human-Machine Interactive Behavior Analysis With Wrist-Worn Devices for Password Inference.

Area of Science:

• Cybersecurity
• Human-Computer Interaction
• Wearable Technology

Background:

• Wearable devices with advanced sensors enable context-aware applications but are vulnerable to side-channel attacks.
• Embedded sensors in smart devices can be exploited by adversaries to infer sensitive user information.

Purpose of the Study:

• To present a self-adaptive and pretraining-independent pattern attack to infer graphical passwords using motion sensor data from wrist-worn smart devices.
• To develop a method for detecting sliding behaviors used to draw graphical passwords and an algorithm for inferring passwords from hand movement trajectories.

Main Methods:

• Developed a novel attack leveraging motion sensor data from smartwatches to reconstruct hand movement trajectories.
• Created a detection method for sliding gestures associated with graphical password input.
• Proposed an inference algorithm to generate password candidates adaptable to various keypad input settings.

Main Results:

• The attack successfully infers complex graphical passwords with up to 80% accuracy in a single attempt.
• Password inference success rates exceed 90% within five attempts.
• Demonstrated the attack's effectiveness without requiring prior victim training data or knowledge of input settings.

Conclusions:

• The proposed attack highlights a significant, overlooked privacy threat posed by sensor data leakage from wearable devices.
• Self-adaptive and pretraining-independent attacks can compromise graphical password security on smartwatches.
• Further research is needed to develop robust defenses against sensor-based side-channel attacks on wearable technology.