At Financial Crime 360 on 2 November 2026, FINTRAIL hosted a roundtable on AI-enabled financial crime risk decisions, led by Francesca Pammer, Associate Director, and Gabriel Dart, Senior Consultant. Financial crime leaders from across the industry joined us to discuss what AI means for accountability, governance and the future role of human judgement. Here is what we heard.
Key takeaways
Explainability is about defensible decisions, not just understanding how a model works.
AI cannot simply be bolted onto existing operating models; governance needs a rethink.
Human roles will shift towards analysis, challenge and judgement, and training must keep pace.
There was broad recognition that AI can significantly improve the speed and efficiency of financial crime operations. But the discussion made clear that successful implementation requires more than introducing new technology into existing frameworks. Firms must rethink how they make, evidence and oversee risk decisions.
Explainability is about defensible decisions, not just understanding the model
The discussion challenged the industry's focus on the technical explainability of AI models. While understanding how a model reaches an output is important, participants questioned whether the same level of scrutiny is consistently applied to more traditional, human-led decisions.
Drawing parallels with established credit-scoring models, the group considered how firms could overlay additional controls, rationale and human judgement onto AI-generated outputs. Combining multiple providers and machine-learning capabilities may also help create a more transparent and explainable process.
Ultimately, the question for firms is whether they can demonstrate to supervisors that decisions are reasonable, evidence-based and appropriate to the risks identified. One useful way to think about an AI agent is as a junior team member: capable of undertaking much of the groundwork, but requiring clear instructions, appropriate supervision and sign-off, with an identified individual remaining accountable for the output. However, the analogy also highlights a resilience challenge that is less pronounced with human staff. If an agent, or the provider supporting it, becomes unavailable, malfunctions, or changes its behaviour, firms must still be able to carry out the underlying activity. That requires tested fallback arrangements, sufficient in-house capability to perform the work without the agent, and a clear understanding of the critical third-party dependencies on which the activity depends. Explainability must go beyond how an algorithm works to include the information/data considered, rationale applied, limitations understood and the governance in place.
A related consideration is whether expectations around decision quality will increase. Because AI can analyse significantly more data points than a human analyst within the same timeframe, firms may be increasingly challenged to explain why relevant information that could readily have been identified by AI was not reflected in a decision. This places greater emphasis on evidencing the data considered and its limitations, not only the outcome.
The group also acknowledged that demonstrating explainability is not an entirely new problem. Firms already have experience overseeing third-party KYC and transaction monitoring solutions, and similar standards of scrutiny and accountability should apply to AI.
AI requires a rethink of governance and operating models
A key takeaway was that firms cannot simply bolt AI onto existing financial crime operating models and expect to realise its full potential.
Traditional structures, including the three lines of defence, were designed around different risks and ways of working. As AI takes on activities traditionally performed by analysts and investigators, firms must question whether these structures remain fit for purpose.
This echoes recent signals from the FCA. In a speech on 22 October 2026, Nikhil Rathi suggested it may be time to reconsider the traditional three lines of defence model. Firms redesigning their operating models around AI therefore have an opportunity to build clearer, less duplicative assurance arrangements, rather than adding another layer of oversight to existing structures.
This means rethinking responsibilities, oversight and decision-making, while taking an ecosystem-wide view of risk that considers critical providers, data protection, operational resilience and interactions between multiple models.
More layers of technology and oversight do not necessarily mean stronger governance if ownership and accountability become less clear.
Human roles and skills must evolve alongside AI
AI is increasingly capable of outperforming humans in information retrieval, document review and summarisation. However, human strengths remain important in assessing plausibility, applying context, challenging outputs and exercising judgement. The objective should be to identify which activities can be handed over to AI and where human intervention genuinely adds value.
Much of this evolution is likely to happen naturally. Financial crime professionals have already adapted to successive waves of technology, from automated sanctions screening to rules-based transaction monitoring and case management systems, and AI is unlikely to be different. As people use AI tools in their day-to-day work, they will learn where the tools perform well, where they fall short and how to get the most out of them. In practice, roles will shift rather than disappear, with less time spent gathering and collating information and more time spent on analysis, challenge and decision-making.
Firms should nonetheless actively support this adaptation, which requires a rethink of training and development. Rather than simply processing cases or deciding whether a payment should proceed, future financial crime professionals will need to interrogate AI outputs, understand data, formulate effective prompts, identify weaknesses and reach defensible conclusions.
There was also concern about the risk of over-reliance on AI and the potential erosion of critical thinking. Using AI to summarise regulatory developments is not a substitute for understanding and interpreting regulation. AI should act as a thought partner, not replace the ability to think critically and exercise independent judgement. Adapting to AI should therefore mean learning to use it well, not learning to depend on it.
What’s next?
AI is already reshaping how financial crime decisions are made, evidenced and overseen. Firms that act now to update their governance, operating models and skills will be best placed to meet supervisory expectations and realise the benefits.
Speak to FINTRAIL. Whether you are piloting AI in screening, monitoring or investigations, or reviewing how your three lines of defence will work alongside it, our team can help you build a framework that is effective and defensible.
Read more. Explore FINTRAIL's thought leadership on AI in financial crime frameworks.
Thank you to everyone who joined us at Financial Crime 360 for such an engaging discussion.