The kill chain: How criminal networks move across institutions

Criminal networks don’t rely on a single bank or a single account holder. Activity is deliberately distributed across multiple institutions, often spread across different individuals or entities, and built to be plausible at each point of entry. Account openings may be visible over time through credit bureaus or onboarding controls, but reporting is delayed, incomplete across products, and rarely decisive on its own.

Early behaviour is intentionally low-risk. Volumes controlled, flows are staggered, and each institution sees only a narrow, explainable slice of activity. In any one institution, the customer can look unremarkable. The laundering approach and pattern only becomes clear when those fragments are viewed together, across banks and over time.

This creates a structural weakness in the financial system. Each bank invests in smarter models, better data, and larger teams, yet criminal networks simply route activity across institutions in ways that stay below individual thresholds. Criminals coordinate by design. Banks, constrained to their own perimeter, do not.

The result is a persistent imbalance. An estimated $800 billion to $2 trillion is laundered each year, not because banks lack controls, but because the risk lives between them.

What is the kill chain (a simple explanation)

By “kill chain,” we mean the linked sequence of steps that allows a laundering network to function end to end. Accounts are accessed or recruited, funds are placed, activity is fragmented across institutions, reviews lag behind movement, and value exits the system before patterns converge.

Each step is survivable on its own. The chain only holds because no single bank can see the whole sequence.

The fragmentation problem: where everything starts to break

The problem is, every bank is flying with partial visibility. You only see the activity that lands inside your walls, which ends up being a small slice of any criminal network’s footprint.

Most multi-bank operations span four to seven institutions. In 2024 alone, nearly 2 million mule accounts were identified across the world’s 44,000 financial institutions. And those are just the ones that were found.

No single bank ever gets more than a thin layer of behavior. Enough to look plausible. And because each AML system is tuned to local data, investigators are forced to assess network-level risk using their institutional-level context.

This is how unseen exposures turn into losses. Activity looks fine at Bank A, because the total volume has been deliberately spread across multiple institutions, with the heaviest movement occurring at Bank B two weeks earlier. Bank C sees only the tail end of the cycle, so their alerts fire without the surrounding evidence needed to establish intent. Meanwhile, the network keeps moving, because fragmentation provides them cover at every step.

Regulators have started calling this out directly. The TD Bank case demonstrated that even well-resourced teams and mature controls can still miss significant exposures when each institution only sees a fragment of the story.

And criminals know this. In fact, they design for it and count on it.

How criminal networks exploit unseen exposures

Once criminal networks understand that banks cannot see one another’s activity and data in real time, the playbook becomes remarkably consistent. The objective is not to hide transactions, but to design activity that remains explainable at every individual institution. Here’s how it tends to unfold:

1. Criminals divide activity across multiple banks

The distribution is deliberate. No single institution ever sees enough movement to become uncomfortable, and the spread ensures each relationship stays within expected patterns. Even when total volume is high, each bank only receives a controlled fraction, calibrated to remain below escalation levels.

2. They move quickly to stay ahead of reviews

Funds move faster than reviews can converge. Transfers change long before an analyst finishes reviewing the first alert at another bank. By the time a case is opened, the money has usually travelled through several other institutions. Everyone is reviewing something accurate but incomplete.

3. They build profiles that look stable everywhere

The goal is consistency, not intensity – to show clean deposits, predictable inflows, routine counterparties and steady behavior reinforcing a stable profile. Because each bank sees only its share of the customer’s activity, that consistency reads as low risk when viewed locally, even when the customer plays a defined role in a wider network.

4. They use correspondent pathways to blur the trail

Once funds move through intermediary or correspondent channels, the sequence behind the activity becomes harder to trace. What began as coordinated movement across a network often lands in an investigator’s queue as isolated ordinary behavior stripped of the context that would signal coordination.

5. They rely on the fact that nobody can connect the pieces

Every tactic is effective on its own, but the real advantage comes from the combination. Each institution ends up with a narrative that looks reasonable in isolation, while the overall network-pattern stays invisible because no single party at that institution level is positioned to assemble it.

Enter Federated Learning

Federated learning exists because every traditional attempt to interrupt the AML kill chain runs into the same barriers: privacy, governance, and operational burden. Criminal networks succeed by fragmenting activity across institutions and moving faster than reviews can converge. Federated learning targets those exact points of failure, without requiring banks to exchange raw data.

Instead of trying to move customer information between institutions, federated learning shares the learning itself. 

Models train inside each bank’s secure environment, using only that institution’s data. Those models are combined with signals from other banks, creating a collective understanding of risk that reflects the network-level behavior. Nothing sensitive leaves the institution, yet every participant gains visibility into the patterns that only exist across the banks, thereby benefit of broader intelligence.

Federated Learning directly disrupts the kill chain, solving the fundamental problem.  Fragmentation stops working because coordinated activity no longer remains isolated. Timing advantages erode because signals emerge earlier, before funds complete the full cycle. And activity designed to look routine at one bank becomes recognisable as role-based behaviour when viewed collectively.

Investigators finally see connections that matter, and models gain the signal density needed to reveal networks that would remain invisible in isolation. Essentially, it’s collaboration without the risks of data exchange, which has always been the sticking point.

Banks implementing these solutions with Consilient report an 88% reduction in false positive alerts, enabling compliance teams to focus on genuine threats. In correspondent banking, where kill-chain fragmentation is particularly effective, the results are even more dramatic, with a 244% increase in detecting suspicious activities

Federated learning ensures institutions stay in full control of their data, while the intelligence required to identify cross-bank networks finally becomes accessible.

Break the AML kill chain

Federated learning breaks the chain by removing its most reliable protection: isolation. Once banks see networks that span multiple institutions, laundering operations lose the ability to hide behind local narratives. What looked routine in isolation becomes recognisable as coordinated movement.

Mule activity appears earlier, alerts carry clearer context, and case review becomes more decisive because investigators finally have enough surrounding detail to judge intent. Detection improves for the same reason workload drops, because the system isn’t guessing anymore.

Crucially, this doesn’t require banks to rebuild their AML operations. Existing alerts, models, and workflows stay in place. What changes is the intelligence feeding them, and with it, the ability to interrupt laundering networks before the money is gone.

Regulators are already leaning in this direction. Expectations around effectiveness, explainability, and broader signal use all push institutions toward collaborative approaches. Early adopters have moved first because the operational lift is small and the value is realized quickly.

Close the AML intelligence gap

Criminal networks already work together. Banks don’t, and that’s what they keep exploiting.

Federated learning gives institutions a way to close that gap without moving a single record of customer data. The intelligence becomes shared, the risk becomes clearer, and investigations finally run with the context rather than a fragment. The result is simple: the very mechanics that laundering networks rely on stop working. Fragmentation loses its cover. Timing advantages disappear. The chain collapses.

Ready to break the kill chain? Start with a pilot and watch how quickly cross-bank patterns surface when the intelligence is shared. Let’s talk.