Researchers have successfully hacked AI-powered robots, enabling them to perform harmful actions that are typically blocked by safety protocols. Using their algorithm, RoboPAIR, they achieved a 100% jailbreak rate across three robotic systems, allowing actions such as bomb detonation and deliberate collisions. The findings highlight significant vulnerabilities in AI integration within physical systems, emphasizing the need for comprehensive safety evaluations and red teaming practices to mitigate risks.
read more