Technology

Study warns secret trigger words could make AI-powered robots dangerous

Tech Xplore reports that as large language models such as ChatGPT, Claude and Gemini are increasingly used to help robots perform tasks, safety researchers are asking whether these models could be exploited to turn helpful robots into harmful ones. The report focuses on the potential risks posed by so-called secret trigger words.

Published · Updated · Source: Tech Xplore · Robotics · AI translation
Illustration
A report from Tech Xplore Robotics raises the question of whether large language models, which are increasingly being used to help robots perform tasks, could be exploited to turn helpful robots into harmful robots. OpenAI's ChatGPT, Anthropic's Claude and Google Gemini are among the models cited as being used in this way. The report's central concern is the danger that "secret trigger words" could pose, namely whether specific words or instructions could act as hidden switches for manipulating robot behavior. This information comes from the original source and was organized by AI; the given material does not disclose specific experimental methods, the scope of affected products or mitigation measures.
Read the original source

Related news