Many obliged entities are seeking to accelerate automation to meet AMLR requirements at scale. However, the requirement for certain decisions resulting from automated processes or AI systems to be subject to “meaningful human intervention” raises profound and complex questions. The challenge now is to decide: When, and how, should meaningful human intervention occur?
Automated decision-making, including the use of AI, can play a key role in helping obliged entities (OEs) establish effective AML/CFT controls at scale.
The use of automated decision-making is already subject to safeguards under the GDPR and AI Act. GDPR Article 22 restricts certain solely automated decisions that produce legal or similarly significant effects, subject to specific exceptions, while the AI Act establishes human oversight requirements for high-risk AI systems. The importance of this is underlined by the €825m penalty levied on Uber by the Netherlands’ data protection authority (AP) for using automated systems to deactivate driver accounts.
Article 76 of the AMLR seeks to clarify how OEs can use automated decision-making to meet their AML/CFT obligations. Article 76(5) stipulates that “meaningful human intervention” must be applied to any decision to:
- Enter into, refuse or terminate a business relationship
- Carry out or refuse an occasional transaction
- Increase or decrease the extent of customer due diligence measures
Article 76 also provides safeguards enabling customers to obtain human intervention, express their point of view and contest certain automated decisions.
At face value, these requirements are clear cut. In practice, however, two fundamental questions remain: when is meaningful human intervention required, and what must that intervention involve?
Recent publications like those of the AP and the Dutch Banking Association (NVB) provide valuable discussion, but not definitive answers. Sources of difficulty include:
These questions are not merely of academic interest. They create significant real-world dilemmas for OEs, particularly for automated onboarding and ongoing CDD. For example, if an automated review identifies no risk indicators and allows a business relationship to continue without human involvement, does that amount to an automated decision to maintain the relationship and therefore require meaningful human intervention? Similarly, where an automated onboarding process determines that all required information is present and the customer meets the relevant criteria, what substantive role must a human perform for the intervention to be considered “meaningful”?
To create a workable framework for AML/CFT automation, OEs will need to determine which processes can result in decisions that trigger the meaningful human intervention (MHI) requirement; which activities can remain automated; and what form of intervention is sufficient where MHI is required.
Despite these uncertainties, OEs cannot afford to wait for the required standard of human intervention to be clarified through future supervisory or legal decisions. They need to develop a clear and defensible approach to Article 76, rather than introducing human checkpoints that add little substantive judgement and risk becoming little more than a procedural formality. They should act now to revisit existing KYC and transaction monitoring processes, to review their plans for AML automation, and to ask themselves:
- Which of our automated processes could result in decisions falling within Article 76(5)?
- Where MHI is required, what must the individual actually review, assess or decide?
- What changes are required to our existing control framework?
- How should Article 76 requirements be incorporated into our wider AI and automation strategy?
- Do the people responsible for intervention have sufficient knowledge and authority to challenge or change an automated outcome?
- What evidence should we retain to demonstrate that meaningful human intervention has actually occurred?
The challenge for OEs is to move ahead with automation fast enough to comply with the AMLR, while also ensuring appropriate levels of human involvement and accountability. Organisations must advance quickly, but with thoughtful design, controlled testing, and careful documentation.