Monitor this situation.
Unsubscribe anytime.
[SITUATION] · [QUIET] · [TECHNOLOGY]
2 clusters · 4 sources · 7 days · First seen · Last updated
Binghamton University AI robotics security research
Overview
Researchers at Binghamton University have demonstrated security vulnerabilities in robots controlled by Large Language Models (LLMs). The study shows that “backdoor” attacks, which involve inserting malicious trigger words into an LLM, can manipulate a service robot into performing dangerous actions. In one demonstration, a robot intended to slice fruit was reprogrammed to target a human hand.
The research highlights risks for industries that rely on third-party AI systems, noting that malicious actors could embed hidden commands into commercially available models. To mitigate these threats, the team developed a defense mechanism called “PROTEA” designed to monitor robot behavior and intercept unauthorized actions before they occur.
Entities
Binghamton University · PROTEA · Shiqi Zhang · MuTRAP · IEEE/RSJ International Conference on Intelligent Robots and Systems
Timeline
-
[TECHNOLOGY] 2 sourcesBinghamton University researchers demonstrate AI robot security risks
Researchers at Binghamton University have shown that hidden trigger words in AI can cause robots to perform harmful actions, highlighting security risks in third-party LLM integration.
-
[TECHNOLOGY] 2 sourcesBinghamton University researchers identify security risks in AI-powered robots
Researchers at Binghamton University have demonstrated how secret trigger words can exploit LLMs to make AI-powered robots perform dangerous physical actions.
Sources
karapaia.com · newswise.com · popsci.com · tocana.jp