Inverted ROC/AUC (0.65) when using sklearn SVM

Question

I fit an rbf SVM to training data and then computed the AUC and plotted an ROC on that data. I expect it to overfit and yet instead the AUC seems to be 0.35. The accuracy is