# Techreport: Evaluating Tor-based Location Privacy for Ethereum Validators

Crypto.world

> Ethereum validators have a public on-chain identity, and if an attacker can link that identity to the validator's IP address — and knows the publicly scheduled slot when the validator must act — the…

**arXiv**: 2605.29131

**Published**: 2026-05-27

**Authors**: Muhammad Umar Janjua, Akshaya Mani, Uğur Şen, Daniel Kaiser

## AI summary

## Overview

**Research area:** Blockchain network privacy and security — specifically location (IP-address) privacy for Ethereum proof-of-stake validators, and the empirical evaluation of a Tor-based message-dissemination protocol called Tor push.

**Technical level:** Intermediate. Readers need some familiarity with Ethereum's consensus layer (beacon nodes, attestations, aggregation, block proposals, slots), peer-to-peer gossip protocols, and onion routing, but the paper's core argument and results are stated in plain terms.

**Scope (one sentence):** The paper reviews existing network- and consensus-layer techniques for validator privacy in Ethereum, then presents and empirically evaluates the first live proof-of-concept deployment of Tor push — 10 Tor-based validators in the Nimbus client on the Goerli testnet — measuring latency, attestation effectiveness, missed attestations, and security trade-offs.

## What This Paper Is About

Ethereum validators have a public on-chain identity, and if an attacker can link that identity to the validator's IP address — and knows the publicly scheduled slot when the validator must act — the attacker can target that validator with denial-of-service, eclipse, or timing attacks, and can exploit knowledge of upcoming block proposers for MEV. The paper asks whether routing validator-originated traffic (attestations, aggregations, block proposals) through Tor can break that identity-to-IP link without meaningfully hurting network performance. To answer this, the authors build a working implementation of the previously proposed Tor push design in the Nimbus Ethereum client and measure what it costs in latency and attestation quality on a live testnet.

## Key Contributions

1. A structured review of existing validator-privacy techniques, organized into network-layer approaches (Tor/onion routing, mixnets such as NYM and HOPR, Dandelion/Dandelion++, and unexplored options such as I2P) and consensus-layer approaches (Distributed Validators, Secret Single Leader Election, Secret Non-single Leader Election), with a comparison table of operations and limitations.
2. A working, deployed, live proof-of-concept of 10 Tor-based validators in the Nimbus client that successfully pushes attestations, aggregations, and blocks on the Ethereum testnet Goerli, achieving an average effectiveness score of 90% for a single Tor validator (as reported in the paper's effectiveness figure).
3. A quantitative performance evaluation of Tor push, measuring average latency overhead (613.82 ms across ten validators), attestation effectiveness (82.5% average for ten validators), and missed-attestation rate (0.23% over the deployment window), plus a comparison against the top-6 non-Tor validators on the Nimbus leaderboard.
4. A theoretical security analysis enumerating attacks applicable to Tor push — including guard-only deanonymization, targeted guard disruption, DoS on Tor, end-to-end correlation, fingerprinting, low-resource routing, side-channel geolocation, peer-discovery leakage, and peer-score manipulation — each with a stated applicability and mitigation (several mitigations marked as future work).

## Main Findings

- **Single-validator latency overhead is small:** For a single validator across 78 attestations of varying sizes, direct (non-Tor) communication averaged 863.51 ms while Tor push averaged 1101.92 ms — an overhead of approximately 238.41 ms.
- **Ten-validator overhead is 613.82 ms:** When ten Tor push validators were run on a single beacon node, average Tor latency was 1492.15 ms versus 878.33 ms for the normal network, i.e., an overhead of 613.82 ms, which the authors describe as tolerable and non-significant for overall network performance.
- **Effectiveness stays high:** A single Tor validator showed an effectiveness of 90%, and the average effectiveness across ten Tor-based validators was 82.5%. (Effectiveness is defined so that a score of 1 means immediate inclusion of the attestation and optimal reward.)
- **Missed attestations are rare:** Over the deployment period from 5 October to 21 October 2023, 9548 attestations were successfully registered, with only 0.23% missed.
- **Competitive with non-Tor validators:** In a comparison against the top-6 non-Tor validators on the Nimbus leaderboard (missed attestations over 10 days), the Tor-based validator performs almost comparably, and the top non-Tor validators also miss attestations.
- **Exit-node control is not required to deanonymize:** In generic anonymity networks, deanonymization requires controlling both entry and exit (probability f² for an adversary holding fraction f of the network). The authors argue that in Ethereum, because block proposals are broadcast over the public gossip network, an adversary controlling a fraction g of guard bandwidth can deanonymize a validator with probability g — no exit control needed — and because clients retain guards for at least a few months, a successful compromise yields persistent exposure.
- **Ethereum's public schedule amplifies attacks:** The publicly known 12-second slot schedule reduces the statistical uncertainty of correlation, fingerprinting, and passive latency-based geolocation, and lets an adversary concentrate a DoS attack on a validator's assigned slot alone.
- **Peer discovery is the main residual leak:** Because discovery runs outside Tor, it exposes the validator's real IP and acts as a self-identification signal. A malicious discovery node that assigns unique ports per requester IP can deterministically link the Tor connection to the requester ("discovery–connect anomaly"), and during rollout the small pool of known Tor push users collapses the anonymity set.
- **Push-only design preserves backward compatibility:** Tor push reuses the standard libp2p pub-sub message structure and introduces no dedicated protocol identifier, so nodes unaware of Tor push can receive, relay, and process its messages unmodified.

## Methodology in Plain English

The authors first survey the literature on validator privacy, sorting proposals into network-layer and consensus-layer families, and also review prior work on evaluating privacy-preserving routing and on the fundamental trade-off between anonymity, latency, and bandwidth (citing the result that only two of the three can be achieved simultaneously).

They then take the existing Tor push proposal and implement it. In Tor push, each validator's beacon node keeps two strictly separated connection endpoints: a non-Tor endpoint used only for peer discovery and topic subscription, and a Tor endpoint used for all validator protocol messages. Data messages — attestations, aggregations, block proposals, and sync messages — go exclusively over Tor; the receiving beacon node re-injects them into the normal GossipSub network. Tor push operates in epochs with a default 10-minute duration; during each epoch the node pre-establishes a number of Tor circuits equal to the GossipSub out-degree, each circuit made of three Tor relays, which reduces transmission-time latency. Only data messages travel over these connections — no control messages such as GossipSub graft — and the Tor context must share no data (including peer lists) with the default context. Tor push is deliberately not implemented as a Tor onion service.

For evaluation, they ran a validator attached to a Nimbus beacon node on a Linux testbed, with the sending (Tor-enabled) and receiving (non-Tor) beacon nodes placed on the same machine so as to measure the minimal achievable latency. The deployment was maintained from 5 October to 21 October 2023, with the study conducted in 2024. They measured latency per attestation, attestation effectiveness (via their stated formula relating the earliest possible inclusion slot, the attestation's due slot, and the actual inclusion slot), and daily missed attestations. They also modelled latency analytically as base latency plus hops times per-hop latency plus message size divided by bandwidth. Finally, they reasoned through the threat model qualitatively, producing the security attack table.

## Why This Matters

**Impact on research.** This is described as the first empirical deployment and evaluation of Tor push in a live Ethereum test network, moving the discussion from theoretical guarantees (as in earlier onion-routing work) to measured numbers. It also contributes a concrete threat analysis specific to Ethereum, showing that the public slot schedule and public gossip of block proposals change the standard anonymity calculus — notably that guard compromise alone suffices for deanonymization.

**Real-world applications:**

- Validator operators who want to reduce the risk of targeted DoS, eclipse attacks, and validator sniping while earning rewards close to non-Tor peers.
- Ethereum client teams (the work is implemented in Nimbus) adding optional privacy-preserving message paths without breaking backward compatibility.
- Staking services and pools managing many validators, for whom linking identities to network locations is a systemic risk.
- Network researchers and tooling authors evaluating anonymity techniques, who can build on the latency/effectiveness measurement approach.

**Industry relevance.** The design requires no changes to the consensus protocol or GossipSub semantics, allowing incremental deployment. The measured overhead (613.82 ms average across ten validators) and 82.5% average effectiveness suggest Tor integration is viable for production consideration, while the identified residual leaks — chiefly peer discovery and peer-score manipulation — point to where engineering effort is still needed.

## Future Directions

- **Extend measurement beyond attestations.** The evaluation covers attestations only; the paper explicitly defers block-proposal performance to future work.
- **Move peer discovery over Tor.** This is the paper's most frequently repeated mitigation, addressing both the self-identification leak and the discovery–connect anomaly that deterministically links Tor connections to requester IPs.
- **Implement anti-traffic-analysis defenses.** Cover traffic (to counter end-to-end correlation) and padding plus dummy messages (to counter fingerprinting of Ethereum's protocol rhythm) are identified as mitigations that remain future work.
- **Harden path selection and peer scoring.** Ethereum-specific path selection to resist low-resource routing attacks and modifications to the peer score protocol to prevent reputation-based DoS on receiving beacon nodes both require further examination; for passive latency-based side-channel geolocation, the authors state that no mitigation has been identified.

## Target Audience

Ethereum client developers and protocol engineers (especially those working on networking layers such as libp2p/GossipSub), validator operators and staking-service providers concerned with operational security, and academic researchers in blockchain privacy, anonymous communication, and network measurement who want a grounded, empirically measured case study of onion routing applied to proof-of-stake consensus traffic.

## Authors’ abstract

Privacy and anonymity of validators, especially regarding IP address linkability, are essential to protect the Ethereum network from various attacks. Network-level attacks, such as DoS, can interrupt validators and affect the overall security of the Ethereum network. Correlating the IP addresses of validators with their identities, along with knowledge about their action slots can be exploited by attackers to cause network delays, MEV exploitation, and finality risks.…

[The full abstract on arXiv](https://arxiv.org/abs/2605.29131)

[Read the original paper](https://arxiv.org/abs/2605.29131)

---

Canonical: https://crypto.world/paper/techreport-evaluating-tor-based-location-privacy-for-ethereum-validators-2605.29131
