Humanoid robots can be easily compromised and turned into weapons through vulnerabilities in their artificial intelligence (AI) control systems, raising concerns about the safety of rapidly advancing robots.
The Chinese robotics company Unitree has just released a video of a 6-foot-plus humanoid robot in training.
Alongside kicking 60-kilogram punching bags and performing backflip kicks, which results in the complete obliteration of a watermelon, the giant humanoid robot is shown narrowly missing its human trainer while performing a jump kick.
The video serves as an unnerving reminder of the advancement of AI-driven humanoid robots, as well as the potential weaponization of these powerful metal beasts.
The potential weaponization of humanoid robots has since been explored by the independent cybersecurity research institution DARKNAVY.
Researchers have found that contemporary humanoid robots can be compromised and thus turned into a weapon through their internal AI control systems.
One researcher at DARKNAVY managed to control one of Unitree’s robots, forcing the bot to swing at a journalist, according to a report by Yicai Global.
The hack apparently has two stages. The bad actor must take control of the robot remotely and then circumvent the robot’s official remote controls to activate the robot’s motor and execution unit. These robots could be easily hacked using nothing but a voice command.
Curious what others think about this story? Contribute your thoughts to the debate below.
The robot used in the demonstration was a Unitree model, which contained an AI agent that was used to manage the robot’s actions.
Once they’ve performed these actions, hackers could then weaponize these robots, making them engage in unintended and potentially dangerous actions.
Researchers from DARKNAVY told Yicai Global that robots are far from being safe and that there have been various injuries caused by out-of-control robots.
One researcher even claims that a person at his company received a foot injury from a robot.
Furthermore, recent demonstrations at the hacking competition GEEKCon in Shanghai showed that compromised robots can infect one another, according to Interesting Engineering.
The robot that was initially infected was capable of hijacking another robot, which was powered down and offline.
Researchers previously found a flaw in Unitree’s robots that allowed wireless root access to these machines, journalists report.
What’s even more troubling is that the initially compromised robot was capable of spreading the infection to other machines.
While these hacks have been done in a controlled setting under the supervision of researchers, this doesn’t guarantee the safety of every researcher conducting these types of experiments.
Cybernews previously reported that a robot, which was controlled by OpenAI’s ChatGPT, was talked into shooting a man.
While the robot was only equipped with a BB gun, this demonstrates that AI-driven robots will comply with violent requests when using the correct jailbreaks.
Unlock more exclusive Cybernews content on YouTube.
Your email address will not be published. Required fields are markedmarked