The phrase ‚humanoid robots as attack vectors‘ just sent a chill up my spine. Though a fix for this specific exploit is reportedly rolling out.
The only thing more haunting than a robot that can hack other robots is a robot that can hack other robots, which can also walk. An author claims that humanoid robots from manufacturer Unitree are vulnerable to an exploit that embeds „themselves on the devices and then infect the next devices in range“, and it’s estimated to affect the entire new generation product line.
This is all according to Andreas Makris, aka Bin4ryDigit on X, and Kevin Finisterre (via Hackaday). They have a GitHub page going over their findings, as well as their attempts to reach out to Unitree. Bin4ryDigit claims UniPwn (the exploit) is „the first public exploit of humanoid robots“.
The GitHub shows that the security handshake made between devices is „laughably simple“, with it just looking for „unitree“ in encrypted packets. This means that, should it be asked to receive data, it checks if the other user is authenticated with a simple string. Once connected, it checks the serial number, initialises the Wi-Fi mode, and sets the country code.
Within this grouping of activities, bad actors can inject payloads, which ultimately allows them to give commands with root privileges, ie taking control.
What makes this specific exploit dangerous is that it can execute commands like injecting malware, and it can even be used to spread to other nearby robots.
