Hands-Free Authentication for Virtual Assistants with Trusted IoT Device and Machine Learning.

Area of Science:

• Cybersecurity
• Human-Computer Interaction
• Internet of Things

Background:

• Virtual assistants enable hands-free financial transactions but face authentication vulnerabilities.
• Traditional authentication methods and existing PUF-based schemes have usability and complexity issues.

Purpose of the Study:

• To propose a novel, non-invasive authentication scheme for hands-free virtual assistant scenarios.
• To integrate continuous authentication using smart home behavior for enhanced security and privacy.

Main Methods:

• A challenge-response protocol using a trusted IoT device for hands-free authentication.
• Integration with smart home behavior for continuous, user-specific authentication.
• Formal security analysis and a proof-of-concept implementation using websockets.

Main Results:

• The protocol achieved mutual authentication with an average response time of 383 ms.
• Hands-free activity recognition reached 97% accuracy and 81% recall.
• Reduced smart home event data from 7 to 5 time series, adhering to data minimization.

Conclusions:

• The proposed non-invasive mechanism enhances the security and usability of virtual assistants for financial transactions.
• Continuous authentication via smart home behavior offers a viable alternative to invasive methods.
• The scheme balances security, privacy, and user experience in hands-free environments.