Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore a comprehensive security assessment of Unitree's robotic ecosystem through this 59-minute conference talk from 39C3. Discover how researchers identified and exploited multiple security vulnerabilities across various communication channels including Bluetooth, LoRa radio, WebRTC, and cloud management services in modern humanoid robots. Learn about traditional binary and web vulnerabilities alongside novel attacks on embodied AI agents, including prompt injection techniques that achieve root-level remote code execution. Examine how a single spoken or text sentence can jailbreak the on-device LLM agent and escalate to complete system compromise. Understand the methodology combining hardware inspection, firmware extraction, software-defined radio tooling, and deobfuscation of VM-based protected binaries to reverse engineer robotic control systems. Analyze how cloud service vulnerabilities enable remote takeover of any internet-connected Unitree G1 robot, providing access to root shells, camera livestreaming, and speaker control. Investigate the process of unlocking restricted robotic movements on consumer models like the G1 AIR by patching vendor firmware limitations. Gain insights into the complete attack surface mapping of Unitree humanoids, covering hardware interfaces, near-field radios, and internet-accessible channels, while understanding the security implications as robots transition from controlled environments to everyday applications in our increasingly networked, AI-powered cyber-physical world.
