Mix Network

The fundamental security property ensuring an adversary cannot correlate outgoing messages with incoming messages. This is achieved through cryptographic transformations and batch mixing.

• Bitwise unlinkability: Even a single bit of difference in a plaintext must produce a completely unrelated ciphertext at the output.
• Traffic analysis resistance: Prevents linking through message size, timing, or volume correlations.
• Formal definition: For any two honest senders, the adversary's advantage in guessing which sent a given message is negligibly better than random.

Each message is wrapped in multiple layers of public-key encryption, forming an 'onion.' Every mix node in the path peels off one layer to reveal routing instructions for the next hop.

• Forward secrecy: Compromise of one node's key does not decrypt the entire path.
• Fixed-size cells: All messages are padded to identical lengths to prevent size-based correlation.
• Replay protection: Unique session identifiers and timestamps prevent an adversary from resending captured messages to observe routing behavior.

A mix node collects a threshold number of messages before processing them as a group. The node cryptographically transforms and randomly permutes the batch, destroying the temporal ordering.

• Synchronous batching: All messages in a round are output simultaneously, preventing timing attacks.
• Stratified mixing: Messages can be grouped by priority or latency class while maintaining anonymity within each stratum.
• Threshold vs. timed release: Nodes either wait for N messages or a maximum timeout T, balancing latency against anonymity set size.

Senders independently select a sequence of mix nodes from a public directory, forming a path of length L. No single node knows both the origin and the final destination.

• Path length trade-off: Longer paths increase latency but improve anonymity against colluding adversaries.
• Free-route topology: Senders choose any sequence, preventing a compromised directory from dictating paths.
• Cascade alternative: Fixed, shared paths where all users route through the same sequence, maximizing anonymity set size at the cost of flexibility.

To resist end-to-end traffic confirmation attacks, mix networks generate decoy traffic. Dummy messages are indistinguishable from real payloads to any observer without the final decryption key.

• Loop cover traffic: Dummy messages that circulate within the network, creating a constant background noise floor.
• Link padding: Continuous transmission of encrypted bits on inter-node links, even when no real messages are queued.
• Statistical uniformity: The inter-arrival time distribution of all messages (real + dummy) follows a Poisson or uniform process, masking bursts of real activity.

While providing anonymity, mix networks can incorporate cryptographic accountability to prevent abuse without deanonymizing honest users.

• Pseudonymous credentials: Users authenticate via long-term 'nyms' that are cryptographically unlinkable to real-world identities.
• Rate-limiting tokens: Anonymous proof-of-work or blind-signature tokens that cap the number of messages a single nym can inject per epoch.
• Selective revocation: A quorum of authorities can revoke a misbehaving nym's anonymity for a specific message if a valid warrant or abuse threshold is met, using threshold decryption.

A rigorous mathematical framework that provides a quantifiable privacy guarantee by injecting calibrated statistical noise into query results or datasets. The privacy loss is bounded by a parameter called epsilon (ε)—lower values mean stronger privacy but reduced accuracy. Key mechanisms include:

• Laplace Mechanism: Adds noise drawn from a Laplace distribution, scaled to query sensitivity
• Gaussian Mechanism: Uses Gaussian noise, providing relaxed (ε, δ)-differential privacy
• Exponential Mechanism: For non-numeric outputs, selects results probabilistically based on a utility score

Mix networks can incorporate differential privacy to protect metadata patterns even if an adversary observes network traffic statistics.

A cryptographic scheme that allows computations to be performed directly on encrypted data without decryption. The result, when decrypted, matches the result of the same operations performed on plaintext. Three levels exist:

• Partially Homomorphic Encryption (PHE): Supports only one operation type (e.g., RSA for multiplication, Paillier for addition)
• Somewhat Homomorphic Encryption (SHE): Supports limited operations before noise overwhelms the ciphertext
• Fully Homomorphic Encryption (FHE): Supports arbitrary computations on ciphertexts, enabling privacy-preserving fraud model inference on encrypted transaction data

While computationally intensive, FHE enables a bank to run fraud detection algorithms on customer data without ever seeing the raw transactions.