We characterize optimal key and message confidentiality under low-entropy keys and construct a distribution-aware encryption scheme that attains the optimal message-recovery bound.

We build a large-scale honey-vault dataset, develop stronger attacks and a Transformer-based decoy model, and evaluate practical security against online guessing.

For HE-based honey vault schemes, we 1) propose a new generic construction and an incremental update mechanism, which resists intersection attacks; 2) formally investigate the optimal strategy for online verifications and further propose several practical attacks, which can effectively distinguish real and decoy vaults for the existing honey vault schemes; 3) instantiate our construction with a well-designed multi-similar-password model, which can generate more plausible-looking decoys.

For existing honey encryption applications, we propose two types of attacks, encoding attacks and distribution difference attacks, and further show the insecurity of the applications. We propose a generic method to transform an arbitrary probability model to a probability model transforming encoder, which resists encoding attacks.