Conducting and Mitigating Portable Thermal Imaging Attacks on User Authentication using AI-driven Methods

Thermal cameras have become portable enough to integrate into wearables, such as glasses, and can be used maliciously to infer passwords observing heat traces left on keyboards, keypads and screens. While prior work showed how AI-driven approaches can be used to further enhance the effectiveness of these attacks, we use similar approaches to detect vulnerable interfaces and obfuscate heat traces to defend against thermal attacks. At our Augmented Humans 2023 demo, attendees will have the chance to use a thermal camera to observe thermal traces on a keyboard, and observe how machine learning can both automatically identify keys pressed based and identify, then obfuscate, thermal images of a keyboard to prevent thermal attacks. This demo will provoke thought and discussion about the security risks presented by discrete, wearable thermal cameras and how these risks can be mitigated by both designers and users.