Privacy in P2P isn't binary. Different systems offer different guarantees:

Confidentiality: message content is encrypted. Most P2P systems provide this via transport encryption (Noise, TLS) and end-to-end encryption (Matrix Olm/Megolm, SSB private-box).

Source anonymity: observers can't determine who originated a message. Provided by Tor circuits.

Destination anonymity: observers can't determine who received a message. Provided by onion services and I2P eepsites.

Unlinkability: observers can't link different actions to the same user. Requires per-session keys, cover traffic, or mixing.

Plausible deniability: you can't prove a node stored or requested specific content. Provided by Freenet and GNUnet.

Tor wraps messages in layers of encryption, each peeled by a successive relay. See Tor and I2P for implementation details.

The client builds a circuit of three relays (guard, middle, exit).

Each relay knows only its predecessor and successor—never both source and destination.

Exit nodes connect to the clearnet; .onion services keep the destination inside the Tor network via rendezvous points.

I2P bundles multiple messages into encrypted "cloves" sent through unidirectional tunnels:

Inbound and outbound tunnels are separate, adding path diversity.

Every I2P node is a router, distributing traffic load.

Designed for internal services (eepsites) rather than exit traffic.

Content encryption protects payload, but metadata (who, when, how much, to whom) is often more revealing. As the saying goes: "We kill people based on metadata."

Cover traffic / padding: send constant-rate dummy messages to mask real activity. Expensive in bandwidth but effective against traffic analysis.

Stream multiplexing: bundle connections to hide individual flows within a larger stream.

Query privacy in DHTs: DHT queries reveal what you're looking for. Techniques include Private Information Retrieval (PIR), two-phase lookups (query via an intermediary), onion-wrapped DHT queries, and GNUnet's R5N.

PubSub subscription privacy: subscribing to a topic reveals your interests to mesh peers. Private PubSub is an open research problem.

.onion / .i2p addresses: self-certifying, no DNS dependency, resistant to censorship.

Petnames: user-assigned local names mapped to cryptographic identifiers. "Alice calls this key 'Bob'."

Self-certifying names: the name is a public key or hash, verifiable without a registry. CIDs, IPNS names, and SSB feed IDs are all self-certifying.

Domain fronting and rendezvous obfuscation: make P2P traffic look like benign HTTPS traffic to CDN domains (where legal and ethical).

Stronger privacy always costs performance:

Source anonymity costs latency (multi-hop circuits add round trips).

Unlinkability costs bandwidth (cover traffic, padding).

Query privacy costs complexity and weakens DHT performance.

Plausible deniability costs storage overhead (storing encrypted content you didn't request).

Applications must choose a point on this spectrum based on their threat model.

Privacy overlays wrap Transport connections in additional encryption layers. They interact with Identity (pseudonymous keys vs real identity; rotating keys for unlinkability). PubSub topic subscriptions leak interest metadata. Content addressing requests reveal what you want. Security threats include traffic correlation attacks that undermine anonymity. Compliance addresses the fundamental tension between anonymity and legal obligations—privacy tools serve both human rights defenders and bad actors. Testing of privacy systems requires specialized metrics (anonymity set size, timing leakage) and privacy-preserving telemetry.