Onion Routing

The core mechanism of onion routing is layered encryption. The originator selects a path through the network and encrypts the message multiple times using the public keys of each successive node in reverse order. The final message is wrapped in the key for the exit node, then that is wrapped in the key for the middle node, and so on. Each router can only decrypt its own layer, revealing only the identity of the next hop and the still-encrypted inner payload. This ensures no single node knows both the origin and the final destination of a message.

Onion routing protocols enforce perfect forward secrecy by generating ephemeral session keys for each circuit. Even if a node's long-term identity key is later compromised, past session keys cannot be decrypted, and historical traffic remains secure. This is achieved through a circuit extension protocol where the client performs a Diffie-Hellman key exchange with each successive hop, ensuring that session keys are never transmitted in a recoverable form. The compromise of any single node's key material does not retroactively deanonymize past circuits.

To thwart traffic analysis, onion routing networks transmit all data in fixed-size cells. In the Tor protocol, for example, every cell is exactly 512 bytes. This uniform packet size prevents an adversary from correlating incoming and outgoing traffic based on message length. Padding cells are generated when no real data is ready to send, maintaining a constant stream of identically sized packets. This defense is critical against traffic confirmation attacks, where an observer tries to link a sender and receiver by analyzing packet timing and volume.

A single onion routing circuit can carry multiple TCP streams simultaneously through a technique called circuit multiplexing. The client assigns a distinct stream identifier to each logical connection, allowing a web browser to load dozens of page elements concurrently over one circuit without building a new path for each request. This reduces latency and minimizes the cryptographic overhead of circuit construction. Streams are independently flow-controlled, ensuring that a stalled download on one stream does not block data delivery on another.

Onion routing networks rely on a set of trusted directory authorities to maintain a consensus document listing all available relays, their public keys, bandwidth capacities, and exit policies. Clients download this signed consensus to construct circuits without needing to discover nodes dynamically. The consensus is updated hourly and requires a majority threshold of authority signatures to be considered valid. This centralized trust model prevents Sybil attacks, where an adversary floods the network with malicious relays to intercept traffic.

Each exit relay publishes an exit policy that defines which destination IP addresses and ports it is willing to connect to. This is a critical security and abuse mitigation feature. A typical policy allows common services like HTTP (port 80) and HTTPS (port 443) while blocking SMTP (port 25) to prevent spam. Clients enforce these policies locally during circuit construction, ensuring that traffic is only routed to exits that explicitly permit the intended destination. This shifts liability management to individual relay operators.

A routing protocol that creates hard-to-trace communications by using a chain of proxy servers. Each server receives messages from multiple senders, shuffles them cryptographically, and sends them back out in random order. Unlike onion routing, mix networks often introduce artificial latency to thwart timing correlation attacks, making them suitable for high-latency, batch-oriented anonymity systems.

A cryptographic protocol enabling multiple parties to jointly compute a function over their private inputs while keeping those inputs completely confidential. In fraud analytics, SMPC allows competing banks to collaboratively train anomaly detection models on their combined transaction data without revealing individual customer records. This extends onion routing's principle of hiding who is communicating to hiding what is being computed.

A mathematical framework that guarantees the output of a data analysis does not reveal whether any single individual's data was included. When combined with onion routing for anonymous data collection, differential privacy adds a provable noise layer that prevents reconstruction attacks. Key parameters include:

• Epsilon (ε): The privacy budget controlling the trade-off between accuracy and confidentiality
• Sensitivity: The maximum impact a single record can have on the query output

A digital signature scheme that allows any member of a defined group to sign a message, proving the signer belongs to the group without revealing their specific identity. This provides transactional anonymity in blockchain-based fraud systems. Unlike onion routing which hides the communication path, ring signatures hide the originator within a crowd, making them complementary for end-to-end anonymous reporting of suspicious activity.

A cryptographic method allowing computations directly on encrypted data, producing an encrypted result that matches operations on plaintext. In the context of onion routing, this enables intermediate relay nodes to perform fraud scoring on encrypted transaction payloads without ever decrypting them. Types include:

• Partially Homomorphic (PHE): Supports only addition or multiplication
• Fully Homomorphic (FHE): Supports arbitrary computation on ciphertexts

A protocol allowing two parties to compute the intersection of their private datasets without revealing any elements outside that intersection. For financial fraud, PSI enables institutions to discover shared blacklisted entities or common fraudulent accounts across their networks without exposing their full customer bases. This extends onion routing's anonymity from communication paths to data ownership.