The Bisq decentralised exchange: on the privacy cost of participation

The Bisq decentralised exchange: on the privacy cost of participation. This 2021 study by Liam Hickey and Martin Harrigan appeared in Blockchain Research and Applications. Bisq lets people exchange Bitcoin for other cryptocurrencies or fiat money without identity checkpoints, but its operation publishes Bitcoin transactions and uses a peer-to-peer network. The study asks what privacy participants give up through those traces. Bisq-specific address clustering can link activity and expose identities, but those links are useful rather than definitive.

Bisq has two components in scope. The Trade Protocol coordinates peer-to-peer exchanges between Bitcoin and other currencies. The DAO, which stands for decentralized autonomous organization, handles governance and finance, using the BSQ token on Bitcoin. A successful trade publishes 4 linked Bitcoin transactions: maker-fee and taker-fee outputs feed a deposit transaction, and that deposit’s multisignature output is later spent by the payout transaction. This distinctive sequence leaves recognizable relationships among inputs and outputs.

The privacy question is not simply whether Bisq asks for a name; it does not enforce identity checkpoints. It is whether an observer can combine the transaction patterns participants necessarily publish with data broadcast across Bisq’s peer-to-peer network, then infer which blockchain addresses belong to the same participant. That matters because grouping addresses can join trading, voting, transfers, and aliases into a richer behavioral record than any single address provides.

Hickey and Harrigan built Bisq-specific clustering rules for Trade Protocol and BSQ transactions. At the time of analysis, Bisq users had completed approximately 90,000 trades. Address clustering means grouping blockchain addresses that are likely controlled by one participant. For trades, the rules follow the trade amount and each participant’s security deposit through the fee, deposit, and payout transactions. For governance activity, they use the structure of BSQ transactions. The researchers then enrich those clusters with data stored in Bisq’s peer-to-peer network.

The method aggregated multiple forms of activity under inferred participants, including trades, votes, and transfers. Hickey and Harrigan found cases where one cluster carried multiple aliases, including combinations of pseudonyms and real-world names. They also concluded that Bisq trades are readily identifiable and that trade amounts and security deposits can be linked to their respective owners before and after a trade. In this evaluated setting, avoiding an identity checkpoint did not prevent activity from becoming linkable.

One concrete reason is transaction flow. Inputs to the maker-fee transaction and the payout output returning the maker’s security deposit can be associated with the maker; the same logic applies to the taker. Because the trade amount changes hands while each security deposit returns to its participant, the researchers can trace the roles through the 4-transaction sequence. Their identified deposit transactions also produced results similar to Bisq’s own identification method, providing a check that the pattern was capturing the intended transaction type.

These mappings are heuristic, not proof of identity. Different labels inside one cluster could indicate a false positive, meaning activity from different participants may have been merged. Hickey and Harrigan inspected the mismatches and judged many to reflect people using several aliases, but that judgment does not eliminate the uncertainty. The method can also miss links. A participant can send BSQ through dummy transfers that make self-controlled movement resemble payment to someone else, shrinking the inferred cluster while adding transaction cost. The analysis also omits details of peer-to-peer messaging, locally stored node data, and the developer ecosystem. Accordingly, it should not be read as a complete account of every possible Bisq privacy exposure.

For privacy engineers and Bisq operators, the exposure comes from recognizable transaction relationships and joining blockchain clusters with peer-to-peer data. The researchers suggest optional dummy-transfer support and user guidance as a defense against the BSQ clustering rule. This can reduce the heuristic’s reach, but it creates otherwise unnecessary transactions and raises user cost. Operationally, detection teams and blockchain analysts should treat clusters as investigative leads rather than verified identities, because both mistakenly merged clusters and deliberately split clusters are possible.

A Bisq-specific method connects recognizable trade and BSQ transaction patterns with peer-to-peer data, revealing linked activity and aliases that can reach real-world names. Identity teams, privacy engineers, exchange operators, and blockchain investigators should account for that linkability when designing or assessing decentralized exchanges. They should not infer that every cluster proves one person, or that a participant who evades this heuristic is untraceable; the demonstrated rules admit both false positives and false negatives.