AI models can perpetuate or even exacerbate bias if they are trained on biased data. For example, if a hiring algorithm is trained on historical data where certain groups were underrepresented, it may unintentionally favor certain candidates over others.