Look Ma, I'm the CEO: Real-Time Video and Audio Deepfakes
Demonstrates real-time executive impersonation using open-source deepfake video and audio models running on commodity GPU hardware.
Summary
The Accessibility of Real-Time Synthetic Media
While generative deepfakes historically required extensive post-processing compute, modern AI frameworks enable consumer GPUs to render high-fidelity synthetic video and voice clones in real time.
Live Executive Impersonation Demonstration
Using publicly available executive keynote videos and podcast interviews, the speaker demonstrates live video and voice transformation during active video conference calls.
Social Engineering Attack Scenarios
Real-time deepfake impersonation defeats traditional voice verification and video identity checks, facilitating high-stakes CEO fraud, password resets, and fraudulent financial transfers.
Detection Frameworks and Defensive Countermeasures
Mitigating live synthetic media attacks requires multi-channel out-of-band verification, biometric liveness detection algorithms, and cryptographically signed video streams.
Related talks
Creator Stage·19 min
Katherine Rackliffe
Presents empirical results from a university research study evaluating user vulnerability and defensive safeguards against AI voice-cloned vishing attacks.
Creator Stage·26 min
Mark Foudy
Demonstrates how deep learning voice cloning bypasses enterprise voice verification systems in realistic adversarial simulations.
Main Stage·23 min
Andrew Logan
Demonstrates how real-time AI voice cloning and software-defined radios can inject fraudulent air traffic control clearances on unencrypted VHF frequencies.