We recently noted that PGP and other e-mail encryption protocols are, in theory, highly vulnerable to chosen-ciphertext attacks in which the recipient of the e-mail acts as an unwitting "decryption oracle". We argued further that such attacks are quite feasible and therefore represent a serious concern. Here, we investigate these claims in more detail by attempting to implement the suggested attacks. On one hand, we are able to successfully implement the described attacks against PGP and GnuPG (two widely-used software packages) in a number of different settings. On the other hand, we show that the attacks largely fail when data is compressed before encryption.
Interestingly, the attacks are unsuccessful for largely fortuitous reasons; resistance to these attacks does not seem due to any conscious effort made to prevent them. Based on our work, we discuss those instances in which chosen-ciphertext attacks do indeed represent an important threat and hence must be taken into account in order to maintain confidentiality. We also recommend changes in the OpenPGP standard to reduce the effectiveness of our attacks in these settings.
Download the paper in PDF format here