Recent research shows the increasing threat to website fingerprints (WF) of privacy-sensitive web users especially with machine learning technique such as deep learning or machine learning(DL/ML) decreased efficiency of previous countermeasures. It caused by the range of features of previous countermeasures manually extract cannot cover the features automatically extracted by DL/ML based attacks. In this paper, we propose a black-box countermeasure to website fingerprint attack based on decision-boundary confusion. It discards the manual selection of features, but uses the classification results of classifiers to determine the decision boundary of classifiers, so as to automatically find the adversarial traffic that can confuse the classifier. At the same time, in order to fix the retrain problem caused by adversarial traffic, we add a method bases on Monte Carlo estimation to confuse decision boundary. Therefore, it is difficult for classifiers to form stable and effective decision boundary after retraining the adversarial traffic. Results shows that our method gets a defense success rate of 72.4% when facing the baseline WF Attacks, outperforming existing SOTA method Walkie-Talkie’s 63.6% defense success rate. At the same time, our method improves the ability of the adversarial traffic to resist retrain, increased the retrain defense success rate from 6.4% to 72.4% under 31% overhead.