'Attackers can poison training data, creating backdoor models that respond to hidden trigger phrases.' Defending against these threats involves advanced detection methods like out-of-distribution probes for harmful input.
#AIsecurity #MachineLearning #Backdoors #DataPoisoning #CyberThreats #shorts
Sign up to get updates from us
