|
MoneroResearch.info |
 |
 |
 |
|
ACK-J, A.-J. 2022. LORD OF THE RINGS: AN EMPIRICAL ANALYSIS OF MONERO’S RING SIGNATURE RESILIENCE TO ARTIFICIALLY INTELLIGENT ATTACKS. Multidisciplinary Academic Grants in Cryptocurrencies. Added by: Jack (20/01/2023, 04:14) Last edited by: Jack (20/01/2023, 04:16) |
 |
| Resource type: Unpublished Work BibTeX citation key: ACKJ2022 View all bibliographic details  |
Categories: Monero-focused Creators: ACK-J Publisher: Multidisciplinary Academic Grants in Cryptocurrencies |
Views: 86/2296
|
Attachments
Lord_of_the_Rings__An_Empirical_Analysis_of_Monero_s_Ring_Signature_Resilience_to_Artificially_Intelligent_Attacks.pdf  [37/892]
|
URLs https://raw.github ... ligent_Attacks.pdf |
| Abstract |
|
Cryptocurrencies such as Bitcoin and Ethereum have seen a rapid increase in consumer adoption over the last decade. However, their lack of privacy guarantees has created a secondary market for more privacy-centric alternatives. Monero is a popular cryptocurrency with $2.9 billion in market capitalization and unique privacy properties which allow users to transact without a discernible history, similar to cash. In a transaction, the sender, receiver, and amount are hidden using well-established cryptographic primitives. The crux of Monero’s strong privacy claims has historically surrounded ring signatures, used to obfus- cate the transaction sender. A few previous works have analyzed the security of Monero’s ring signature implementation, but none have assessed its updated on-chain resiliency to AI-based attacks. In this work, we develop a process to collect large-scale datasets composed of Monero transactions accompanied by ground truth labels. Using this process, we built two datasets from the Monero testing and staging networks and used them to explore feature engineering and model selection. These datasets are used to train various supervised-learning classifiers, simulating an adversary who aims to remove the anonymity set of a Monero ring signature. Our most effective classifiers achieve a weighted F1-score of 34.60%, predicting an out-of-sample subset, and a macro F1-score of 13.30%, predicting against real mainnet Monero transactions. The model predictions show a marginal 4.30% increase in accuracy compared to the random guessing probability of 9%. Our research found that there to be minimal transaction risk posed by on-chain information leakage, correlated with adjacent Monero blockchains. We hope this work facilitates future multifaceted research into strengthening the Monero protocol against attacks correlating side-channel information.
|
| Notes |
|
https://github.com/ACK-J/Monero-Dataset-Pipeline
Conclusion: "The key cryptographic primitive, ring signatures, of the privacy-centric cryptocurrency Monero, has been tested by limited previous works against AI attacks. This work contributed a first-of-its-kind pipeline to produce de-anonymized datasets of Monero transactions. The two datasets published represent various user spending patterns. One mimics the expected distribution of the blockchain, while the other represents users who spend Monero as fast as possible. Additionally, we explored the effectiveness of two machine learning models and 19 one deep learning model in the task to identify the true spend of an arbitrary Monero ring signature, absent external information. Our best performing models achieved accuracies upwards of 34.60% predicting on an out-of-sample subset of the dataset and 13.30% on real mainnet transactions. Compared to the 9% accuracy achieved by random guessing, the Monero protocol, as implemented, has shown significant resilience to on-chain information leakage. We hope our open-source datasets and collection pipeline enable future works to test the weaknesses of Monero against various adversarial scenarios." Added by: Jack Last edited by: Jack |