How the Tor Protocol's Three-Node Circuit Isolates Identity from Destination

When a user opens the Tor Browser and navigates to a website, their internet service provider sees only an encrypted connection to a random server, while the destination website sees traffic arriving from an entirely different machine. The link between the human at the keyboard and the server hosting the content is mathematically severed.[2]

This separation is not achieved by trusting a commercial virtual private network to keep no logs. Instead, it relies on "onion routing," a protocol originally developed by the US Naval Research Laboratory in the 1990s and detailed in the seminal 2004 USENIX paper by computer scientists Roger Dingledine, Nick Mathewson, and Paul Syverson.[1][4]

The architecture forces every piece of data through a circuit of exactly three volunteer-operated servers, known as relays. "The design of Tor is intended to prevent attackers from linking communication partners, or from linking multiple communications to or from a single user," the 2004 USENIX paper states.[1]

To understand the isolation, consider the encryption process before the data ever leaves the user's device. The Tor client software selects three relays from a directory of approximately 8,000 global nodes: an Entry Guard, a Middle relay, and an Exit relay.[2][6]

The client then encrypts the data three times, utilizing 128-bit AES encryption and 1024-bit RSA keys for identity verification, using the public keys of the chosen relays in reverse order. The innermost layer is encrypted for the Exit relay, the middle layer for the Middle relay, and the outermost layer for the Entry Guard.[1][5]

This nested encryption resembles the layers of an onion. When the packet, fixed at exactly 512 bytes to prevent traffic analysis based on size, leaves the user's computer, it travels to the Entry Guard.[1][7]

The Entry Guard is the only node in the circuit that knows the user's actual IP address. However, because the data is still wrapped in two layers of encryption, the Entry Guard cannot read the payload or know the final destination. It only knows it must forward the 512-byte cell to the Middle relay.[4][5]

Upon receiving the packet, the Middle relay strips off its layer of encryption. The Middle relay knows the identity of the Entry Guard and the Exit relay, but it is entirely blind to both the user's IP address and the website being accessed. It serves as the cryptographic air gap.[2][6]

Finally, the packet arrives at the Exit relay. The Exit relay removes the final layer of encryption, revealing the actual destination, such as a web server. The Exit relay sends the traffic to the open internet.[5]

Crucially, the Exit relay knows what data is being requested and where it is going, but it has no idea who requested it. The traffic appears to originate from the Exit relay's own IP address.[2][4]

"Tor provides a distributed network of servers... so no single point can link you to your destination," the Electronic Frontier Foundation explained in its 2012 anonymity guide.[3]

This three-node architecture fundamentally shifts the trust model. A traditional VPN requires absolute trust in the provider, who can see both the user and the destination. Tor requires only that at least one of the three relays in the circuit is honest and not colluding with the others.[1][6]

The protocol also rotates these circuits automatically. By default, the Tor client builds a new circuit every 10 minutes for new connections, meaning a user browsing for an hour will route traffic through six entirely different paths. This limits the window of exposure if a circuit happens to be compromised by an adversary controlling multiple nodes.[2]

However, this privacy comes with a strict performance penalty. Because every 512-byte cell must be encrypted three times, routed across the globe to three different servers, and decrypted at each hop, the latency is inherently high.[1][7]

Furthermore, the Exit relay represents a vulnerability if the underlying traffic is not independently encrypted. If a user connects to an unencrypted HTTP website, the Exit relay can read the plaintext data, even though it does not know the user's identity.[3][5]

The protocol's reliance on volunteer relays means its capacity is permanently constrained by the bandwidth those volunteers donate. The mathematical guarantee of anonymity holds firm, but it dictates that the network will always trade speed for the certainty that identity and destination remain isolated.[2][7]