The Expert View: How AI is reshaping financial crime, risk and compliance decision-making at scale

The promise of generative AI as a tool for investigating financial crime

 

As generative AI promises to remake entire professions, it’s still unclear what the new technology will mean for financial crime. The wide availability of image-generation tools has made the materials for deep-fake identity fraud widely available, but it’s also revealed new possibilities for tracking and investigating financial crime at scale.

 

Agentic AI firm Silent Eight brought together financial crime and compliance professionals from some of the world’s largest financial organisations for a free-ranging conversation about how generative AI is changing their work. With budgets shrinking and criminals finding new avenues for money-laundering, AML and compliance teams are in dire need of new tools for spotting financial crime. The only question is how to make AI-powered systems reliable enough for deployment.

 

First and foremost, compliance professionals are looking to AI as a way to manage pressure on their bottom lines. Across the industry, teams are expected to do more with less – and the anticipation of AI-powered efficiency gains has often made that pressure more intense. One attendee said that 20 per cent of his organisation’s employees had been reassigned on the expectation that AI tools would make their work more efficient, leaving the remaining employees scrambling for tools that could fill the gap. Even when resources are available, organisations are often more comfortable allocating the budget towards software spending, rather than committing to additional full-time employees.

 

Another source of pressure comes from regulators themselves, who no longer see it as sufficient to check clients against a list of published names. OFAC’s 50 per cent rule – which requires banks to bar any entity 50 per cent or more of which is owned by a blocked person – now explicitly includes indirect ownership and control, which makes the task of compliance far more complex. Teams are now expected to actively seek out their clients’ indirect partners, checking each new name against an ever-expanding watchlist. The war in Ukraine, which sparked an explosion of financial crime and money laundering, has raised the stakes still further.

 

AI has already proven to be a crucial tool in managing this complexity. In particular, attendees were optimistic about the use of LLMs to detect subtle patterns across large sets of data. One gave the example of regular transactions from a mobile software business: each individual transaction might look legitimate, but tracking the overall flow of currency would show that the funds were being redirected from a gambling business. It’s the kind of distributed warning sign that human workers rarely have the chance to assess – but a properly trained AI model might easily spot.

 

In other cases, the model enabled banks to introduce unstructured data that would previously have been unworkable. Another attendee gave the example of checking video feeds from local bank branches, verifying that a client had actually been present when the reported transaction took place. It’s the kind of extra-mile verification that would have been impossible to scale without generative AI tools, which can now scan through the footage and verify the transactions automatically.

 

Attendees were sceptical about whether contemporary AI tools could master the subtlety that experienced officers are able to bring to their work – but the tools may still be able to provide value without that. Even if AI systems are limited to collating data and flagging inconsistencies, it would free up significant time for compliance officers, leaving them with more hours in the day to put their hard-won wisdom to work.

 

Regulatory oversight also presents a challenge when it comes to improving the model. Most AI models run on a loop of continuous feedback and improvement, but the highly regulated nature of financial crime and AML work make this approach uniquely difficult. Because models are closely scrutinised for fairness and efficacy, any updates have to go through the same slow approval process. The result is that, even when the models are learning from experience, the improvements often can’t be automatically applied.

 

Unfortunately, financial criminals are continuously improving too. As one attendee noted, a criminal who gets turned away once is likely to try again with a more complex laundering scheme. The result is that, as banks get better at spotting bad money, their job only grows more complex. Generative AI tools can offer the chance to finally pull ahead in that ongoing arms race – a prospect that the industry as a whole seems to anticipate with optimism.

---

To learn more, please visit: www.silenteight.com