No! That is the beauty of this algorithm. It is extremely robust to overfitting - training more models on subsets of the training set and taking their mean does not lead to overfitting but generally delivers improved performance (beyond a certain point the gains become very small or cease alltogether).