Sophisticated money launderers know how to scatter funds across institutions and hide behind multiple ultimate beneficial owners (UBOs). The Anti-Money Laundering Council (AMLC) receives a very large number of covered and suspicious transactions reports from financial institutions each year.
Yet, it cannot keep pace with complex illicit fund flows. Recent examples include vast cash movements tied to Philippine Offshore Gaming Operators (POGOs) and suspicious real-estate investments by politically exposed persons.
By using multiple financial institutions and multiple UBOs, sophisticated launderers ensure that each institution sees only a fragment of the flow. A single bank’s compliance team may never connect the dots because their view is limited to accounts within their system. Only the AMLC, with its centralized pool of suspicious and covered transaction reports, is in a position to run cross-institutional analysis and spot the networks hidden across sectors.
Without a functioning AI/ML engine to stitch these reports together, these complex schemes will continue to exploit blind spots.
The question is whether the AMLC’s new procurement of an AI/ML-powered detection engine truly equips it to close this gap.
The ideal AI/ML detection engine would not stop at looking for exact matches in names or account numbers. It would take every suspicious transaction report and look for the less obvious links: a phone number that reappears under a different name, an address written slightly differently, or a counterparty that shows up across several banks.
Instead of treating these details separately, the system would build them into a connected map of accounts, people, and institutions. From that map, it could highlight clusters of activity that resemble laundering rings, identify accounts that act as key conduits, and follow money as it moves quickly from one institution to another.
On top of this, the ideal AI/ML detection engine would assign risk scores and anomaly detection to bring unusual behavior to the surface. Even the narrative fields in reports, often written as short notes, could be mined with natural language processing to pull out hidden names or details. In practice, this kind of engine would allow AMLC to see patterns and stories hidden in the noise.
AI/ML detection is supposed to replace the existing rules-based monitoring system. Rules-based models rely on thresholds and static red flags. Criminals have long learned to stay just below these thresholds or spread their activity across banks and accounts. An AI/ML-driven approach does not rely on those rigid boundaries.
Instead, it learns patterns from large volumes of data, identifies anomalies, and builds risk scores dynamically. It allows AMLC to detect laundering even when no single transaction breaks a rule, because the pattern of transactions as a whole still raises suspicion.
The procurement record of AMLC’s AI/ML detection engine deserves closer attention. The TOR requires the winning bidder to provide a complete AI solution: delivery, setup, installation, ingestion of historical STR/CTR data, testing, and user training for one year. It also specifies that the system must be capable of generating a completeness index for STR narratives using natural language processing, and of scoring incoming reports for risk to assist with triage.
The budget was set at ₱66.05 million, with ₱60 million allocated for the AI software and ₱6.05 million for training AMLC personnel. In December 2022, after earlier failed bids, the contract was awarded to Micro-D International, Inc. for implementation within 400 days from the Notice to Proceed.
There has been little public reporting on whether procurement milestones have been met.
By AMLC’s own disclosures, the project is not a one-off; around ₱50 million annually is needed for subscription and maintenance, effectively locking the agency into a long-term relationship with the vendor. For an agency that often struggles to secure adequate funding, this is not a minor concern. This also raises questions about vendor dependence, value-for-money over time, and whether AMLC negotiated sufficient flexibility to adapt or upgrade the platform as laundering methods evolve.
Even if the technology functions as promised, there are deeper issues. The value of machine learning depends on the quality of the inputs. STRs are only as useful as the data reported by banks and other covered institutions. If fields are left blank, inconsistent, or deliberately vague, no algorithm will be able to connect the dots.
Another gap is operational. A detection engine can generate risk scores and network alerts, but it cannot replace investigative judgment. With only a limited number of analysts, the question remains whether AMLC has the capacity to act on the intelligence the system will produce.
The project is a necessary leap forward, but it should be seen as a first step. Technology can expand visibility and highlight hidden structures in financial flows, but AMLC and covered institutions will still need better reporting discipline.
The author is a lawyer and founder of the law firm Geronimo Law and financial advisory firm Strago. He is a professor of law and finance, author, and athlete.
