A Bloom filter is a probabilistic hash-based data structure extensively used insoftware products including online security applications. This paper raises the following importantquestion: Are Bloom filters correctly designed in a security context ? The answer is no andthe reasons are multiple: bad choices of parameters, lack of adversary models and misused hashfunctions. Indeed, developers truncate cryptographic digests without a second thought on thesecurity implications.This work constructs adversary models for Bloom filters and illustrates attacks on three applica-tions, namely Scrapy web spider, Bitly Dablooms spam filter and Squid cache proxy. Conse-quently, the adversary forces the filter to systematically exhibit worst-case behavior. One of thereasons being that Bloom filter parameters are always computed in the average case. We com-pute the worst-case parameters in adversarial settings, and show how to securely and efficientlyuse cryptographic hash functions. Finally, we propose several countermeasures to mitigate ourattacks.