Original title: Population-Calibrated Graph Screening at 835-Million-Address Scale, with Label-Free Transfer to New Chains
Yury Korolev · September 2026
Checking crypto wallets for illegal activity usually means matching them against public banned lists. This fails when bad actors use new, unlisted addresses or switch to blockchains with no lists at all. To fix this, researchers built a system that scores wallets by tracking how they connect to others in a giant transaction network, rather than looking for names on a list. It maps patterns across five blockchains using a shared model. Because it learns general network behavior, it can spot risky accounts on entirely new chains without needing training lists for them. It sets alert cutoffs using exact slices of the whole user base, so analysts know how many warnings to expect. The system caught dangerous wallets long before public lists named them, worked quickly in live testing, and revealed specific blind spots when tested against simulated attackers designed to trick it.
Imagine you're at a crowded event and need to spot suspicious people without a guest list. Normally, you'd only recognize them if their face is on a known wanted poster. Instead, you watch who talks to whom: someone chatting with groups that have been in trouble before stands out, even if you've never seen them. In this paper, the system treats blockchain addresses like event attendees, scoring them by their transaction links, so it can flag risky ones on new blockchains without any pre-existing labels.
Compliance teams can spot risky blockchain addresses years before public registries, preventing transactions that traditional methods miss and protecting users and regulators.
The abstract does not specify the details of the blind spot exposed by the adversarial harness. The system is tested only on five EVM chains, so its performance on non-EVM blockchains is unknown. It controls alert volume through quantile thresholds, but real-world evasion strategies beyond the tested archetypes may not be addressed.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Population-Calibrated Graph Screening at 835-Million-Address Scale, with Label-Free Transfer to New Chains — Yury Korolev
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.