SoK: Cryptocurrency Mixing and Anonymity - Architectures, Threat Models, Operational Aspects and Security

This survey provides a comprehensive review of mixing proposals and existing implementations, beginning by summarizing a set of review criteria for mixing services, focusing on control structures, obfuscation primitives, and robustness. Subsequently, it systematically analyzed the proposed systems and explored exemplary attack vectors. Furthermore, it provided a detailed comparison of the services and highlighted the threats and limitations inherent to their architectures.

The review reveals that the mixing ecosystem is diverse and filled with innovative approaches. However, few of the existing solutions have successfully achieved a completely undetectable transfer of capital. This challenge remains highly debated and is further compounded by increasing regulatory scrutiny and governmental backlash. Addressing these limitations will require new paradigms that balance user privacy, system efficiency, and regulatory compliance. Future work should explore advanced cryptographic techniques, such as multiparty computation and zero-knowledge proofs, alongside decentralized governance models.

The paper reviews mixing schemes based on classifications such as:

(de-)centralization:

"Centralized mixing service is an unintuitive approach to fund transfer obfuscation in an otherwise decentralized network. The advantage of having a centralized mixing body is reduced network communication overhead and, therefore, increased efficiency and coordination [4, 3]. They provide an easily expandable and straightforward interface at the cost of a centralized nature. A centralized institution is in its entirety trust-based, meaning that users rely on the reputation and goodwill of service providers not to be rid of their capital [3]. Another thing to mention is the opaqueness of a centralized mixer, where such a service may keep track of user information."

"Decentralized mixers meet the vision of a decentralized service – a key concept for blockchains. A direct consequence can be reduced operation speed and difficulties with scalability. Although not suffering from a centralized trust-based architecture prone to scamming, decentralized mixing services can also suffer from certain attacks. An example of such an attack is a DoS-type attack in which an attacker enters the mixing pool along with other users. The attacker then refuses to sign the transaction, resulting in the inability to execute mixing (famous examples are CoinJoin-based systems [17]). Another possible attack (Sybil attack [38]) can happen when the adversary controls a majority or a large number of transactions in the mixing pool. This allows them to better track and deanonymize other users, rendering the mixing process ineffective."

"Centralized cross-chain exchanges Wu et al. identified another supplementary class of mixing – cross-chain services [2, 39]. These services further impede taint analysis by transferring funds between cryptocurrencies. They require extensive synchronization and robustness in order to facilitate safe and lossless exchange."

(Inner Processes/Obfuscation Primitives):

"An important grouping feature in mixers is the principle of obfuscation. The ecosystem of existing techniques is vast; therefore, we focused on the approaches and methods used by the services described in the included papers and practical services (see Sec. 4). The considered obfuscation techniques involve: swapping (e.g., [17, 18]), shuffling (e.g., [19, 20]), aggregation (e.g., [9, 10, 21]), peeling chains (e.g., [9, 10]), randomized fees (e.g., [10, 22, 23, 24, 25, 21, 26, 27]), randomized delays (e.g., [10, 22, 17, 19, 20, [etc.]), address freshness (e.g., [etc.]), off-chain transactions (e.g., [9, 10, 23]), zero-knowledge proofs (e.g., [24, 25, 27]), other techniques, such as cross-chain transactions (e.g., [28, 29, 30, 31]), ring signatures (e.g., [26, 23]), TEEs (e.g., [22, 32, 33, 34, 35], etc.)"

(Proposals vs. Implementations):

"Finally, we can divide approaches by their academic nature vs. the real-world implementation: This results into the following categories: 1. peer-reviewed publications from academia without any implementation or with a proof-of-concept implementation; 2. peer-reviewed publications from academia with full implementation; 3. fully operational implementation, potentially with a white-paper."

(Specific Approaches and Mechanisms):

"Helix. Based on the approaches outlined above, we can characterize Helix as a service using several privacy measures. Based on the analyses of Balthasar [11] and Möser [8], it is clear that Helix utilized a form of centralized swapping.

Here are specific improvements to AI systems based on the insights from this survey, categorized by application:

)1. Enhanced Anonymity and Privacy in Decentralized Finance (DeFi) Applications:

The paper highlights a diverse landscape of mixing services (CoinJoin, CoinShuffle, CoinParty) and privacy-preserving cryptocurrencies (Monero, Zcash). The key takeaway is that anonymity is contingent on the size of the anonymity set and the effectiveness of the underlying cryptographic primitives.

Improvements for AI Systems:

• Develop an AI agent capable of dynamically selecting or constructing optimal mixing protocols based on real-time network conditions (e.g., transaction volume, block confirmation times) to maximize anonymity while minimizing latency (addressing User-added delay).

• Create a machine learning model trained to predict the vulnerability of specific decentralized mixers (e.g., CoinJoin variants) against Sybil attacks or DoS attacks based on observed network topology and participant behavior.

• Design AI agents that utilize Zero-Knowledge Proofs (ZKPs) like those in MixEth or Tornado Cash, allowing them to verify capital ownership and execute complex financial transactions without revealing the underlying sensitive data (sender, receiver, amount).

Improved AI System Capability:

This system can act as a secure DeFi intermediary or transaction orchestrator. It could facilitate high-value transfers between users on a public blockchain while ensuring that the transaction flow is obfuscated using dynamically chosen mixing techniques (e.g., switching from CoinJoin to CoinShuffle if certain network conditions are detected) and cryptographically proven ownership, thereby maximizing user privacy against taint analysis and deanonymization attempts.

)2. Advanced Threat Detection and Forensics in Blockchain Environments:

The survey details various deanonymization techniques, including taint analysis, temporal analysis, and machine learning models used by chainalysis. It also discusses the vulnerabilities of TEEs (Intel SGX) to side-channel attacks and rollback attacks.

Improvements for AI Systems:

• Build a sophisticated AI forensic tool that integrates temporal correlation analysis with transaction pattern recognition to proactively identify suspicious tainted flows, even when standard taint analysis fails due to low volume or fixed fees.

• Develop an AI model specifically designed to detect subtle timing anomalies indicative of Ring Signature exploitation in privacy coins like Monero by analyzing block time variance and transaction frequency patterns across a network.

• Create a vulnerability scanning AI that simulates side-channel attacks against TEE implementations (like SGX) by analyzing instruction patterns or memory access sequences to predict potential data leaks.

Improved AI System Capability:

This system serves as an autonomous blockchain security auditor and investigative tool. It can move beyond simple signature checks to detect sophisticated, low-level deanonymization attempts—such as timing attacks on Monero transactions or hardware-level exploits against TEEs—by learning the subtle statistical patterns that human investigators or simpler algorithms miss.

)3. Secure Cross-Chain and Interoperable Systems:

The paper reviews centralized cross-chain exchanges and privacy-preserving cryptocurrencies (Oasis, Secret) that use TEEs for confidential smart contracts. It also notes the need for robust synchronization in cross-chain services to impede taint analysis.

Improvements for AI Systems:

• Design an AI agent focused on optimizing cross-chain transaction paths between different Layer 1 and Layer 2 networks by predicting which chain/protocol offers the best combination of anonymity set size and low latency, minimizing synchronization overhead.

• Develop a system that uses ZKPs (as in MixEth) to verify the integrity of data transferred across disparate blockchain environments without needing to trust the intermediary exchange infrastructure itself.

Improved AI System Capability:

This AI can function as an intelligent interoperability layer for decentralized applications (dApps). It would autonomously route complex, multi-step transactions across various blockchains, selecting the most privacy-preserving and efficient path available at that moment, ensuring transaction integrity is maintained through ZKPs rather than relying solely on centralized exchange trust.

Public blockchains record transaction histories that enable address clustering, taint analysis, and cross-service attribution, thereby motivating the development of mixers and privacy layers. Our work presents a structured scoping review of 22 representative systems, defining a common unlinkability objective and five adversary archetypes. We evaluate these systems against a taxonomy of attack surfaces, including chain analysis, timing inference, custodial compromise, coordination abuse, network metadata, and trusted execution compromise. While nominal anonymity-set size and cryptographic strength characterize privacy in theory, effective anonymity in practice depends on transaction denominations, cover traffic, relayer behavior, and compliance-interface design. Distinguishing nominal from effective anonymity, we derive four core lessons: (1) Privacy is strongest when integrated into everyday transactions, since standalone mixing creates an easily profiled user subset; (2) Trust points, including operators, peer quorums, and hardware enclaves, must be explicit so users know who can break privacy; (3) Network metadata, including gas funding and timing, must be treated formally as protocol data in privacy evaluations; and (4) Compliance should use auditable cryptographic predicates for selective disclosure rather than broad operator discretion. Ultimately, our systematization clarifies the strengths, failures, and future requirements of blockchain privacy architectures.