Graph Hacks · Track 1 — Agents that act on connected data
Find the mule ring.
Prove the smallest freeze.
An AI agent investigates AML alerts on a FalkorDB transaction graph — rings, money trails, laundering capacity — then proves which accounts to freeze on a copy of the graph before anyone acts. Every claim is a graph query you can replay.
- can reach H3
- $30,100can reach H3
- accounts in the ring
- 7accounts in the ring
- freezes stop it
- 2freezes stop it
Case board
Four scripted cases — two suspicious, two honest false alarms
Burst of transfers from A17
14 transfers from account A17 within 2 hours to 12 different recipients, most of them opened in the last 30 days.
Open investigationUnauthorized $9,800 debit from C42
Customer complaint: $9,800 left account C42 without authorization on Oct 14.
Open investigationTop-1% anomaly score: M8
Merchant account M8 is in the top 1% of the rules-engine anomaly score (high volume, high degree).
Open investigationShared-device cluster around A95
Four accounts (A95-A98) log in from the same tablet and IP address.
Open investigationWhy the graph is load-bearing
Every step the agent takes is a native graph operation.
Ring detection
Accounts tied together by shared devices and IPs, then measured for pass-through behaviour — algo.WCC
Money trails
Shortest routes between accounts, kept only when real transfers follow them in time order — algo.SPpaths
Shared entities
Multi-hop pivots that link accounts through one device or IP — Cypher patterns
Minimum freeze, proven
The smallest set of accounts that cuts every route — verified on a copy of the graph — algo.maxFlow + minimum cut