ARGOS: Tracing Suspicious Blockchain Transactions
An open-source platform, built in France, for finding and tracing suspicious transactions across blockchains at the scale regulators actually need.
- Funder
- Bpifrance
- Programme
- i-Démo (France 2030)
- Partners
- Functori, LMF, Université Paris-Saclay, LIP6, Sorbonne University
The challenge
Cryptocurrency is used for money laundering, extortion, fraud and the funnelling of money to terrorist organisations. The properties that make the blockchain attractive to everyone else make this hard to police: it is decentralised, it moves fast, and it has no central authority to ask.
Traditional finance handles this through identification, authentication and document checks carried out by an institution. A blockchain validates the exchange itself rather than the people making it. Safeguards do exist, and police and the courts can sometimes connect a transaction to a person, but not at the scale or speed the problem demands.
Tools for this do exist. Most are not built for the European financial system, and most are closed, which is an awkward basis for evidence. France has its own regulatory framework under PACTE, and digital asset service providers have concrete obligations under it: authenticating clients, freezing accounts, identifying addresses. They need tooling that matches those obligations.
The approach
ARGOS stands for Analyse et Représentation des Graphes des Opérations Suspectes, the analysis and graphical representation of suspicious operations. Tarides is building it with Functori, the LMF laboratory at Université Paris-Saclay and the LIP6 laboratory at Sorbonne University.
The platform is designed around a few commitments that shape the engineering:
It is sovereign. ARGOS is designed and developed in France, to European practice and regulation, so European users do not have to hand sensitive financial information to parties outside Europe to analyse it.
It is open source. The platform is published under an open-source licence, source and tests included. Transparency is not a nice-to-have here. If an analysis is going to support a regulatory or legal decision, the method behind it has to be open to scrutiny.
It has to cope with blockchain-sized data. Blockchain histories are large and keep growing, so a meaningful part of the work is research into large-scale data management, which is close to the problems Tarides already works on in storage and in OCaml performance.
It traces rather than flags. ARGOS applies taint analysis and pattern detection to follow assets as they move, analyses smart contracts to understand nested chains of interaction, and follows transactions across bridges between chains, so an asset is not lost at the point it changes network. Machine learning is used to improve the quality of what comes back.
Finally, it has to be usable. A digital asset service provider should be able to run ARGOS alongside the monitoring tools they already have, with an interface they can adapt and reports suited to their own obligations rather than someone else's.
Where the project is now
ARGOS runs to 2028, and we are working with our partners on the research and engineering described above. We wrote the project up when it began, in this earlier post introducing ARGOS, and we will publish more as the work produces results worth showing.