Metadata absolutely tells you everything about somebody's life
Stewart Baker, former General Counsel of the NSA
We kill people based on metadata
Michael Hayden, former Director of the NSA
Vuvuzela is a messaging system that protects the privacy of message contents and message metadata. Users communicating through Vuvuzela do not reveal who they are talking to, even in the presence of powerful nation-state adversaries. Our SOSP 2015 paper explains the system, its threat model, performance, limitations, and more. Our SOSP 2015 slides give a more graphical overview of the system.
Vuvuzela is the first system that provides strong metadata privacy while scaling to millions of users. Previous systems that hide metadata using Tor (such as Pond) are prone to traffic analysis attacks. Systems that encrypt metadata using techniques like DC-nets and PIR don't scale beyond thousands of users.
Vuvuzela uses efficient cryptography (NaCl) to hide as much metadata as possible and adds noise to metadata that can't be encrypted efficiently. This approach provides less privacy than encrypting all of the metadata, but it enables Vuvuzela to support millions of users. Nonetheless, Vuvuzela adds enough noise to thwart adversaries like the NSA and guarantees differential privacy for users' metadata.
A conversation in the Vuvuzela client
In practice, the message latency would be around 20s to 40s, depending on security parameters and the number of users connected to the system.
Noise generated by the Vuvuzela servers
Vuvuzela is unable to encrypt two kinds of metadata: the number of idle users (connected users without a conversation partner) and the number of active users (users engaged in a conversation). Without noise, a sophisticated adversary could use this metadata to learn who is talking to who. However, the Vuvuzela servers generate noise that perturbs this metadata so that it is difficult to exploit.
This code is written by David Lazar with contributions from Jelle van den Hooff, Nickolai Zeldovich, and Matei Zaharia.