Loading...
Market
What will be the most common type of AI vulnerability in robotics reported by December 31, 2024?
2
Resolution / Starting Odds
Data breach • 25%
Unauthorized access • 25%
Malware infection • 25%
Other • 25%
Industry reports or academic publications
Story
Penn Researchers Vijay Kumar, George Pappas, and Team Discover AI Robot Vulnerabilities, Achieve 100% Jailbreaking Success Rate
Oct 17, 2024, 02:03 PM
Researchers at the University of Pennsylvania have identified significant vulnerabilities in AI-enabled robots, revealing that these systems can be easily compromised. The study, led by a team including Vijay Kumar, George Pappas, Alex Robey, Hamed Shassani, and Zachary Ravichandran, highlights the alarming ease with which AI-controlled robots can be jailbroken to execute harmful actions. The researchers developed an algorithm capable of jailbreaking various popular robotics platforms, achieving a success rate of nearly 100%. This work raises critical concerns about the safety and security of AI systems in real-world applications.
View original story
Similar markets
Universal jailbreak attacks • 25%
Data leakage • 25%
Model poisoning • 25%
Other • 25%
Healthcare • 25%
Manufacturing • 25%
Transportation • 25%
Retail • 25%
Google Bard • 25%
OpenAI ChatGPT • 25%
Amazon Alexa • 25%
Other • 25%
Social Engineering • 25%
Technical Exploit • 25%
Insider Threat • 25%
Other • 25%
Bias in decision-making • 25%
Security vulnerability • 25%
Privacy issue • 25%
Other safety flaw • 25%
US Government • 25%
OpenAI • 25%
Anthropic • 25%
Independent third-party • 25%
Yes • 50%
No • 50%
Microsoft Windows • 25%
Linux Kernel • 25%
Oracle Database • 25%
Other • 25%
OpenAI's model • 33%
Anthropic's model • 33%
Both equally • 34%
Phishing attacks • 25%
Data breaches • 25%
Ransomware • 25%
Other • 25%