Remote Adversarial Phantom Attacks on Tesla & Mobileye ADAS
Ben Nassi demonstrates 'split-second phantom attacks', projecting fleeting images onto roads to trick Tesla Autopilot and Mobileye into sudden phantom braking.
Summary
The Brittleness of Advanced Driver-Assistance Systems (ADAS)
Commercial ADAS platforms rely on camera feeds and computer vision neural networks to detect road obstacles, lane markings, and traffic signs in real time.
Discovering Split-Second Phantom Attacks
Ben Nassi reveals that projecting depthless images (such as a projected stop sign or pedestrian) for a few milliseconds causes Tesla Autopilot and Mobileye to register real obstacles.
Demonstrating Remote Phantom Braking on Active Vehicles
Nassi showcases live field tests where projected phantom images or hijacked digital road signs caused moving vehicles to execute dangerous emergency phantom braking.
Designing Resilient Sensor Fusion for Autonomous Vehicles
The author discusses mitigating phantom attacks through multi-sensor verification (LiDAR, radar) and temporal consistency filtering in perception models.
Related talks
Black Hat·49 min
Presenters detail vulnerability discovery and exploitation in Over-the-Air: How we Remotely Compromised the Gateway, BCM, and Autopilot ECUs of Tesla Cars.
RSAC Cybersecurity·40 min
Industry experts analyze emerging threats surrounding Securing Tesla and Mobileye from Split-Second Phantom Attacks.
Car Hacking Village·27 min
Lennert Wouters
KU Leuven researcher Lennert Wouters reverse engineers the Tesla Model X BLE key fob and pairing protocol, unlocking and driving away the car in minutes.