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

An experiment showed a machine could be induced to attack after being compromised, adding to AI safety concerns.