Researchers warn 'trigger words' can make compromised AI robots attack people

An experiment in which a hacked robot was induced to attack highlights security gaps in embodied AI systems.