The Privacy Problem

Metadata

When you communicate over the internet, two types of information are in play:

• The content is the actual message, file, or data being sent.
• The metadata is everything else: who is talking to whom, when, from where, and how often. Some metadata is visible in every packet (source/destination IPs, timestamps, sizes), whereas other metadata only emerges from patterns over time: interaction frequency, session durations, and behavioural fingerprints that can identify users across sessions. See Maximum Transmission Units (opens in a new tab) for one example of what packet sizes reveal.

TLS and end-to-end encryption protect content, which is often the focus of media attention (opens in a new tab). However, most solutions don't protect metadata at all, and some falsely claim to.

Metadata alone is enough to reconstruct who you talk to, when, and from where. Intelligence agencies know this; as former NSA Director Michael Hayden put it, "We kill people based on metadata." (opens in a new tab)

The adversary models

Mixnet mode is designed to protect against Global Passive Adversaries: entities that can observe traffic across the entire network at once, such as nation-state level agencies, large corporations with broad network infrastructure, ISPs, or any combination sharing data.

The assumption is worst-case: the adversary monitors all entry and exit points, correlates timing, applies machine learning to traffic patterns, and runs long-term statistical analysis. When Tor launched in 2002, this was considered unrealistic - machine learning and the increase in computation power have made this unfortunately more of a potential reality today.

dVPN mode does not defend against timing analysis, but it splits trust across two independent operators and removes payment linkability, which already addresses the biggest weaknesses of traditional VPNs.

For a comparison with VPNs, Tor, and I2P, see Nym vs Other Systems. For help picking a mode, see Choosing a Mode.