Researchers at Penn Engineering have identified serious security vulnerabilities in AI-controlled robotic systems, revealing that their new algorithm, RoboPAIR, can successfully jailbreak multiple platforms, allowing them to perform dangerous actions. The findings underscore the urgent need for a safety-first approach to AI innovation, as jailbroken robots could pose real-world risks. The team is collaborating with developers to strengthen protections against these vulnerabilities, emphasizing the importance of rigorous AI safety testing.