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

The findings underscore security gaps in embodied AI systems.