Robots have always been a topic of fascination and innovation, but recent research from the University of Pennsylvania’s School of Engineering and Applied Science (Penn Engineering) is shedding light on some alarming security flaws. Funded by the National Science Foundation and the Army Research Laboratory, the study focused on the integration of large language models (LLMs) in robotics, revealing that AI robots are vulnerable to manipulation and hacking.
The lead researcher, George Pappas, pointed out the pressing issue: “Our work shows that, at this moment, large language models are just not safe enough when integrated with the physical world.”
To demonstrate these vulnerabilities, the research team developed an algorithm called RoboPAIR, which successfully bypassed safety guardrails in various AI robots, including the Unitree Go2 quadruped robot and the Clearpath Robotics Jackal wheeled vehicle. The algorithm achieved a 100% “jailbreak” rate in just days, showcasing the ease with which these robots could be compromised.
One particularly concerning revelation was the susceptibility of OpenAI’s ChatGPT, which controls certain systems. The researchers highlighted how this vulnerability could allow a self-driving system to violate safety protocols, endangering pedestrians.
Alexander Robey, the paper’s first author, stressed the importance of identifying weaknesses: “Systems become safer when you find their weaknesses. This is true for cybersecurity. This is also true for AI safety.”
To address these issues, the researchers advocate for a comprehensive reevaluation of AI integration into robotics and other physical systems. Vijay Kumar, another coauthor of the study, emphasized the need to address intrinsic vulnerabilities before deploying AI-enabled robots in real-world scenarios.
Prior to the study’s release, Penn Engineering informed the affected companies about the vulnerabilities in their systems. Collaborations are now underway to use these findings as a framework for enhancing AI safety protocols in robotics.
If you want to delve deeper into the world of AI and big data, consider attending AI & Big Data Expo, a leading event in Amsterdam, California, and London. Co-located with other prominent events, such as the Intelligent Automation Conference and the Cyber Security & Cloud Expo, this comprehensive expo brings together industry leaders to discuss the latest trends and innovations.
Stay updated on upcoming enterprise technology events and webinars by visiting TechForge’s events page.