AI in Transaction Monitoring: Theory, Practice and … Tango

1. The Theory: Wolfsberg’s Call for Responsible Innovation and Agile Governance

Artificial Intelligence has long been positioned as the future of transaction monitoring, a tool capable of cutting through the noise of false positives and detecting suspicious activity with greater accuracy.

The Wolfsberg Group’s second statement on effective monitoring for suspicious activity, released in August 2025, reinforces this direction.

Wolfsberg calls on financial institutions to move beyond outdated rules-based approaches and embrace AI, machine learning, and automation in their monitoring frameworks.

Key themes from Wolfsberg’s statement include:

• Simplified governance: Wolfsberg highlights inefficiencies in current governance structures, where overlapping validations, assurance, and audit functions delay the rollout of new models. This siloed approach means it can take months—rather than weeks—to implement tools designed to detect emerging risks. To address this, Wolfsberg advocates for streamlined, non-redundant validation frameworks across the three lines of defence, ensuring speed and agility without sacrificing oversight.
• Transparency and explainability: Financial institutions must ensure that AI-driven models are interpretable and can be explained to both regulators and internal stakeholders.
• Continuous improvement: Models should be rigorously tested, trained on real-world cases and improved over time to keep pace with evolving typologies.
• AI as an enabler, not a replacement: Technology should support, not substitute, the judgment of skilled AML investigators.

In short, Wolfsberg provides a vision: a future where AI enhances monitoring effectiveness, supported by governance frameworks that are both robust and agile.

2. The Practice: Bunq Bank’s Regulatory Reality

Just days before Wolfsberg’s latest statement, De Nederlandsche Bank (DNB) announced a €2.6 million fine against Bunq Bank for serious AML deficiencies. The contrast between Wolfsberg’s theoretical framework and Bunq’s regulatory outcome is sparkling … and an insight into the regulators’ eyes.

Between January 2021 and May 2022, Bunq failed to adequately investigate suspicious transaction alerts and maintain sufficient ongoing monitoring of certain customers.

From the detailed report (which can be found published online), many of these deficiencies pointed to weaknesses in AI-driven monitoring processes:

• Alerts failed to capture obvious discrepancies between customer-facing information and actual transaction activity.
• Investigative dispositions merely repeated transaction data, without meaningful analysis or red flags.
• Open-source intelligence checks were missing, raising questions about the legitimacy of high-value flows.
• The outputs of alert handling appeared templated—more aligned with untrained AI outputs than with human investigative reasoning.

The regulator’s message was clear: innovation cannot come at the expense of governance and … the AI tool is only as good as the people who trained it … (and further monitored and reviewed it).

While Wolfsberg advocates for agility, the National Bank’s actions demonstrates that, in practice, regulators demand solid, demonstrable governance, backed by rigorous oversight and human intelligence.

3. The Tango – Between Theory and Practice

The tango between Wolfsberg’s guidance and Bunq’s fine (both published in the same week) captures the delicate balance that banks must … dance.

On one hand, global standards encourage agility, innovation and the responsible use of AI.

On the other, regulators expect thorough, well-documented and human-centered monitoring processes—even if that means slower implementation.

The path forward lies in reconciling these two perspectives:

• Use AI to support, not replace, human investigators.
• Build governance frameworks that do not suffocate AI agility, but are still strong and efficient – no room for superficiality here.
• Train AI models on practical, case-driven data while maintaining transparency and accountability.
• Ensure audit and validation functions are aligned, avoiding redundancy but preserving rigor.

Conclusion

AI in transaction monitoring is not a choice between speed and control—it is about dancing the right balance.

Wolfsberg’s vision of agile, simplified governance is forward-looking and necessary to keep pace with financial crime.

Yet, Bunq’s penalty is a sobering reminder that regulators remain uncompromising when it comes to oversight.

The lesson for financial institutions is clear: the future of monitoring lies in responsibly dancing between advanced technology and solid, demonstrable governance, ensuring that agility never comes at the expense of compliance.

Welcome to the tango!

By Andreea Tampu, ACAMS, ACSS

All articles published on this website are property of the author and may not be duplicated, copied, modified or used in any way without my written permission.

The use of this website does not constitute any right or license to use my articles and presentations without my prior written permission.