Some humanoid robots can be hacked through voice commands, demonstration shows

Researchers are making progress in “terrible robot apocalypse” section.

Yicaiglobal: reports (via Interesting engineering) from Chinese cybersecurity research group Darknavy recently demonstrated a method to compromise commercial humanoid robots that has somewhat dire consequences if used in the real world. At GEEKcon in Shanghai, researchers took a commercially available Unitree robot and demonstrated that it can be hacked with voice commands and, if that’s not bad enough, used to infect other nearby units.

The robot in question escaped from the internal AI agent. By taking advantage of a software flaw, the researchers were able to take over the robot while it was connected to a network, at which point the researchers instructed the robot to use a local wireless connection to spread the hack to another nearby robot that wasn’t actually even connected to the network at the time. It only took a few minutes for the hack to spread from one robot to another. Even worse, the researchers were able to command the robot to physically hit a mannequin on stage.

The implications here are as obvious as they are dire. For decades, cyber security risks have mostly been about stealing money or information from people. If humanoid robots become commonplace in homes (for use in elderly care or the like), that suddenly opens up a whole new world of terrifying possibilities for hackers to legitimately cause physical harm to innocent people. Beyond that, it could disrupt robotic work operations or critical infrastructure.

For now, let’s just be happy that this was all done in a safe environment, with the goal of finding ways to solve these problems rather than spreading them.