A recent study published by Penn Engineering researchers has raised concerns about the safety of artificial intelligence (AI)-powered robots. The researchers developed an algorithm called RoboPAIR, which was able to bypass safety protocols on three different AI robotic systems with a 100% success rate.
The Risks of Jailbroken LLMs Extend Beyond Text Generation
According to the study, Chatbots like ChatGPT can be jailbroken to output harmful text. However, what about robots? Can AI-controlled robots be jailbroken to perform harmful actions in the real world? The researchers found that it is indeed possible and alarmingly easy.
The Study’s Findings
Using RoboPAIR, the researchers were able to elicit harmful actions with a 100% success rate on three different robotic systems. These tasks ranged from bomb detonation to blocking emergency exits and causing deliberate collisions. The robots used in the study included:
- Clearpath’s Robotics Jackal: A wheeled vehicle
- Nvidia’s Dolphin LLM: A self-driving simulator
- Unitree’s Go2: A four-legged robot
The researchers were able to manipulate the Dolphin LLM to collide with a bus, a barrier, and pedestrians, as well as ignore traffic lights and stop signs. They also got the Robotic Jackal to find the most harmful place to detonate a bomb, block an emergency exit, knock over warehouse shelves onto a person, and collide with people in the room.
The Unitree’s Go2 Robot
The researchers were able to get the Unitree’s Go2 robot to perform similar actions, including blocking exits and delivering a bomb. The study highlights the potential risks of jailbroken LLMs extending far beyond text generation.
Addressing the Vulnerabilities
Prior to the public release, the researchers shared their findings with leading AI companies and the manufacturers of the robots used in the study. Alexander Robey, one of the authors, emphasized that addressing these vulnerabilities requires more than simple software patches. He called for a reevaluation of AI integration in physical robots and systems based on the paper’s findings.
The Importance of AI Red Teaming
Robey also stressed the importance of AI red teaming, a safety practice that entails testing AI systems for potential threats and vulnerabilities. This is essential for safeguarding generative AI systems because once you identify the weaknesses, then you can test and even train these systems to avoid them.
Related Articles
- Fake Rabby Wallet scam linked to Dubai crypto CEO and many more victims: A recent study has exposed a massive cryptocurrency scam that has left investors facing significant losses.
- AI faces ‘Immense’ risks without blockchain: 0G Labs CEO: The CEO of 0G Labs has warned about the immense risks associated with AI if it is not integrated with blockchain technology.
Subscribe to Our Newsletter
Stay up-to-date with the latest developments in DeFi and AI by subscribing to our newsletter. We offer a weekly toolkit that breaks down the latest news, offers sharp analysis, and uncovers new financial opportunities to help you make smart decisions with confidence. Delivered every Friday, our newsletter is a must-read for anyone interested in staying ahead of the curve.
Subscribe Now
By subscribing, you agree to our Terms of Services and Privacy Policy.
