When “hidden” isn’t really hidden: study shows privacy risks in virtual meeting tools
Two contributions from the Machine Learning and Security (MLSec) group at BIFOLD have been accepted for presentation at the 34th USENIX Security Symposium (USENIX 2025). The event will take place from August 13 to 15, 2025, in Seattle, USA.
One of the featured studies, “Seeing Through: Analyzing and Attacking Virtual Backgrounds in Video Calls”, reveals that popular video meeting apps don’t always protect user privacy as well as they might assume. The team found that virtual backgrounds can leak small fragments of the real environment during a call. By piecing together these tiny “leftover” pixels, an attacker could reconstruct parts of a caller’s actual surroundings. The paper will be presented by Felix Weissberg.
USENIX Security is a premier conference where researchers, engineers, and security experts share the latest findings on protecting computer systems and networks.
Research Paper
- Seeing through: Analyzing and Attacking Virtual Backgrounds in Video Calls. Felix Weißberg, Jan-Malte Hilgefort, Steve Grogorick, Daniel Arp, Thorsten Eisenhofer, Martin Eisemann, Konrad Rieck.
- Prompt Obfuscation for Large Language Models. David Pape, Sina Mavali, Thorsten Eisenhofer, Lea Schönherr.