Skip to content
Crypto.world

Research

Deanonymizing Monero Transactions in Tor Network

Monero routes peer-to-peer traffic through Tor and I2P so that an observer cannot link a transaction to the IP address that created it. The authors show that Monero's Tor integration creates a…

Deanonymizing Monero Transactions in Tor Network
arXiv
2607.07062
Published
2026-07-08
Authors
Ruisheng Shi, Shihan Zhang, Yulian Ge, Lina Lan, Qingfeng Zhang, Qin Wang

AI summary

Overview

Research area: Computer security and privacy — specifically network-layer deanonymization of privacy-focused cryptocurrencies (cs.CR), combining Monero P2P protocol analysis with Tor traffic analysis.

Technical level: Advanced. The paper assumes familiarity with Dandelion++ transaction propagation, Monero peer management (graylists, whitelists, Timed Sync messages, block-height thresholds), and Tor circuit and hidden-service mechanics.

Scope: The paper presents and empirically evaluates ProxyMark, a three-stage framework that identifies the role of a Monero node operating over Tor, captures the transactions that node originates, and links a Tor-level identifier to the node's real source IP address.

What This Paper Is About

Monero routes peer-to-peer traffic through Tor and I2P so that an observer cannot link a transaction to the IP address that created it. The authors show that Monero's Tor integration creates a structural weak point: a Monero Tor node sends its own transactions only to two outgoing Tor hidden-service peers (called proxy nodes) before those transactions reach the clearnet, whereas relayed transactions go straight to ordinary neighbors. The goal is to exploit that asymmetry — occupy those outgoing hidden-service connections, get adversarial nodes selected as proxy nodes, and then bind the resulting Tor-side identifier to a real IP address using traffic watermarking, all without breaking Tor encryption or Monero cryptography.

Key Contributions

  1. ProxyMark, a three-stage deanonymization framework for the Monero Tor network, chaining node role identification, originated transaction identification, and node location deanonymization into a single pipeline rather than treating them as independent attacks.

  2. An onion address analysis method that distinguishes Monero Tor hidden service nodes from Tor client nodes by exploiting a deterministic difference in Timed Sync Response peer lists: hidden service nodes append their own onion address as the last entry, while client nodes only forward learned addresses.

  3. An originated transaction capture method combining connection occupation (filling target graylists and whitelists with malicious onion addresses so targets reconnect to adversary nodes) with proxy selection bias (advertising falsified, fresher block heights so occupied peers pass Monero's proxy-selection filter).

  4. A role-specific watermarking method that embeds identifiers into Monero P2P message timing so a malicious Tor entry relay can recover the identifier and bind it to the source IP, with handshake-free embedding for hidden service targets and noise cancellation for client targets.

  5. Empirical validation on the live Tor network, Monero mainnet, and testnet, plus a root-cause analysis and a discussion of protocol-level mitigations for each attack stage (onion-address advertisement, transaction forwarding over anonymity networks, peer selection, and message-rate handling).

Main Findings

  • Role identification is exact in the reported test: Onion Address Analysis achieved 100% precision and recall on the Monero testnet. The classification rule inspects the last onion address in three consecutive Timed Sync Responses; identical across all three indicates a hidden service node, and the repeated address is recorded as its onion address.

  • Outgoing connections can be broadly occupied: Connection Occupation captured 7–11 of 12 outbound hidden-service connections after restarts and 8–10 of 10 after periodic replacement, in a controlled live Tor hidden-service deployment.

  • Falsified block heights meaningfully shift proxy selection: With one occupied connection, proxy selection bias raised the proxy-selection rate from 15.3% to 35.7%, a gain of +20.4 percentage points, measured on the Monero mainnet.

  • Watermarking recovers identifiers with high accuracy: Role-Specific Watermarking reached 100% precision with 93.8% recall for hidden service nodes and 100% precision with 91.4% recall for Tor client nodes, using Monero testnet nodes over the live Tor network.

  • The vulnerability is structural, not cryptographic: The attack does not break Tor encryption or Monero cryptography. It exploits the boundary where Monero's forwarding policy assigns different roles to public and hidden-service connections, so that a transaction arriving over a hidden-service link from a proxy peer is, by Monero's own rules, originated by that peer.

  • Protocol details enable the bias: Monero Tor nodes do not validate block height information in Timed Sync messages, and hidden-service peers update height only through once-per-minute Timed Sync exchanges (block sync happens only with public peers), creating a window in which falsified heights are believed.

  • Graylist filling is cheap: Because each Timed Sync Response can carry up to 250 onion addresses and the graylist holds 5,000 entries with FIFO eviction, 20 coordinated malicious nodes suffice to fill a target's graylist, with responses delayed inside Monero's 120-second timeout window.

  • Two protection strategies exist and both have limits: Originated transactions are padded to fixed 3072-byte white-noise segments sent every 10 to 15 seconds, and proxy nodes apply a Poisson-distributed delay averaging 22.5 seconds before clearnet forwarding. The paper notes Monero Tor nodes may disable these protections, in which case originated transactions appear in the mempool immediately via fluff-phase broadcast to all outgoing hidden-service peers.

Methodology in Plain English

The authors first dissect how Monero nodes behave when their P2P traffic runs over Tor, separating ordinary nodes, Tor client nodes (public-only, or with hidden-service peers), and Tor hidden service nodes. From this they derive three observations: peer exchange reveals node role, originated transactions take a proxy-only path, and the Monero layer and Tor layer each see only half of the picture.

The attack then runs in three steps. First, the adversary runs a modified Monero Tor hidden service node, accepts incoming Tor connections, and repeatedly sends Timed Sync Requests. Because a hidden service node always lists its own onion address last, while a client node's last entry changes each time, three responses are enough to classify the peer and, for hidden services, extract the onion address.

Second, the adversary gets its nodes into the target's outgoing connection set. For hidden service targets, it sends Timed Sync Responses stuffed with malicious onion addresses directly. For client targets, which cannot be dialed inbound, it fills the graylists of hidden service nodes first and lets normal peer exchange carry the malicious addresses onward. When the target drops and replaces outgoing connections, it draws replacements from those poisoned lists. The adversary also runs modified nodes that skip full blockchain sync, learn the current height from ordinary nodes, and advertise current height plus a small offset (for example, +5 blocks) every 20 seconds instead of once per minute, so its peers look fresher and survive the proxy-selection height filter.

Third, the adversary embeds a node identifier into traffic timing. For hidden services the identifier is the onion address, encoded as an index; for clients it concatenates a malicious-node index and a per-connection index. Timed Sync Request messages act as carriers because Monero does not rate-limit them: three carriers in a 6-second window mean a 1 bit, one carrier means a 0 bit, and two initialization windows of five (hidden service) or six (client) carriers mark the circuit. A malicious Tor entry relay recognizes the initialization pattern among inbound RELAY cells, decodes the identifier, and attaches the source IP it already observes. For hidden service targets the adversary injects the watermark before completing the Monero handshake (Monero allows up to 256 KB pre-handshake), which sidesteps white noise and response obligations; for client targets, protocol-required noise messages are deferred into signal windows to cancel their effect.

Why This Matters

Impact on research. Prior network-layer deanonymization work targets Bitcoin's diffusion propagation. Because Monero uses Dandelion++ with a stem phase, those methods do not transfer, and this paper supplies a Monero-specific attack surface plus a measured baseline for how much leakage the Tor integration introduces. It also reframes anonymity-network integration as a source of new, role-dependent identifiers and peer lists rather than a pure privacy gain.

Real-world applications (defensive and analytical):

  • Informing Monero protocol changes to onion-address advertisement, transaction forwarding over anonymity networks, peer selection, and message-rate handling — the mitigation areas the authors enumerate.
  • Giving blockchain-analytics and law-enforcement investigators a documented, measurable technique for attributing privacy-coin transactions, relevant given Monero's described role in darknet markets.
  • Helping anonymity-network designers assess how application-layer peering behavior can leak through encrypted transports.
  • Supporting exchange and custody operators in assessing whether Tor-routed Monero nodes expose origin information.

Industry relevance. The results quantify the cost of the attack in ordinary resources: 20 coordinated nodes to fill a graylist, 5,000 onion addresses for graylist population, and adversary Tor relays. That places the barrier within reach of modestly resourced actors and makes the findings directly actionable for wallet and node operators deciding whether to run Monero over Tor.

Future Directions

  1. Protocol-level mitigations. The authors state that they analyze root causes and discuss mitigations across four areas — onion-address advertisement, transaction forwarding over anonymity networks, peer selection, and message-rate handling — but do not report measured effectiveness for any of them. Evaluating those fixes is the obvious next step.

  2. Reducing reliance on a single classification rule. The paper's primary rule requires three consecutive Timed Sync Responses and defers a decision otherwise. An appendix variant based on the last_seen timestamp is mentioned but not evaluated in the reported experiments; comparing single-response and repeated-response classification under adversarial conditions remains open.

  3. Countering the noise-cancellation dependency for client targets. Client-node watermarking needs the target to first connect outbound to a malicious hidden service node, and its recall (91.4%) trails the hidden service case (93.8%). Improving detection under Tor flow-control cells and Monero white noise for clients is an open engineering question.

  4. Generalization beyond the tested parameters. The identifier encoding is sized for 179 measured active hidden service nodes (8 bits) and for a 5,000-node/116-connection client parameterization (13+8 = 21 bits). How the scheme scales and how detection degrades as the Monero Tor population and connection counts change is not reported.

Target Audience

Researchers and graduate students in blockchain security, traffic analysis, and anonymity systems; Monero and Tor protocol developers evaluating privacy guarantees; and security engineers at organizations that route cryptocurrency P2P traffic through anonymity networks. The paper is not beginner-friendly — it presumes working knowledge of Monero's peer management and Tor circuit internals — but its three-stage structure makes the attack logic legible to intermediate readers who already understand cryptocurrency networking.

Authors’ abstract

Monero is a privacy-focused cryptocurrency that deploys the Dandelion++ protocol and incorporates anonymity networks (such as Tor and I2P) to prevent malicious attackers from linking transactions with their source IPs. In this paper, we demonstrate that Monero's integration of the Tor network introduces a fundamental vulnerability: a Monero Tor node's originated transactions are exclusively forwarded to two outgoing Tor hidden service nodes (proxy nodes) prior to clearnet…

The full abstract on arXiv
Read the original paper