Paper 2025/997 Original: eprint.iacr.org/2025/997 ↗

Kerblam — Anonymous Messaging System Protecting Both Senders and Recipients

Abstract

While popular messaging apps already offer end-to-end confidentially, end-to-end metadata privacy is still far from being practical. Although several meta-data hiding systems have been developed and some like Tor have been popular, the proposed solutions lack in one or more aspects: the Tor network is prone to easy low-resourced attacks, and most others solely focus on anonymity for senders or receivers but do not both. Some recent solutions do consider end-to-end anonymity, however, they put significant restrictions on how users use the system. Particularly, the receivers must stay online or trust online servers that receive messages on behalf of receivers. This work presents a scalable end-to-end anonymity messaging system, $\mathsf{ORAM}^{-}$, that overcomes the mentioned issues and restrictions. It stems from a key observation that combining the recently-emerged oblivious message retrieval (OMR) primitive with oblivious shuffling can offer the desired end-to-end anonymity without severely restricting the number of messages a sender may send or a receiver may receive. We build our solution using two non-colluding servers and recent OMR protocol HomeRun and a compatible oblivious shuffle protocol. We then extend our solution to allow larger messages by employing a novel two-server distributed oblivious RAM technique, called $\mathsf{ORAM}^{-}$. Our performance analysis demonstrates that with the increase in the number and size of messages, the performance improvement brought by $\mathsf{ORAM}^{-}$ becomes higher. Specifically, for $2^{20}$ messages of size 1KB, our scheme only needs $5.577$ s to transmit a message.