Exacerbating Algorithmic Bias through Fairness Attacks