SIBOS 2023: 金融犯罪防止におけるAIの影響度を測定する
Key Sibos Takeways for Risk
After nearly a year at the helm of Celent's Risk practice, I had the privilege of attending my inaugural Sibos conference. My colleagues Patricia Hines, Gareth Lodge and Colin Kerr have already blogged about their insights from Sibos on payments, digital transformation, and trade finance, so here are some key observations for our Risk clients.
Embracing Advanced AI for Risk
The first theme that stood out from the conference was the use of more sophisticated forms of AI within the Risk and Compliance function. Two examples of this featured prominently at the conference:
- GenAI Fever. I can not overstate the ubiquity of Generative AI at this conference. Attendees were eager to understand how banks are leveraging GenAI within their operations, the new risks it introduces, and how vendors are integrating GenAI into their solutions. Accenture estimates that 55% of banks are presently evaluating or testing GenAI, of which 23% plan to incorporate it into their 2023/4 roadmap. Notably, the risk function is at the forefront of this trend, with some of the most extensively tested GenAI use cases. These include using GenAI as a co-pilot for AML or Fraud investigators, analyzing and consolidating beneficial owner information, and generating narratives for alerts and SARs.
- The Adoption of Machine Learning. The public's embrace of GenAI has accelerated the adoption of Artificial Intelligence (AI) in all its forms. One of the clearest example is shift from rule-based behavior detection in Transaction Monitoring to a more pure Machine Learning (ML) approach. While several solution providers already use ML to reduce false positives generated by rule-based AI, Google's entry into the AFC (Anti-Financial Crime) technology arena with their AML AI product, along with HSBC's application of that product, has given this approach substantial credibility. This has made a market that other vendors are keen to step into. Google was there demoing its new product but so were its regtech competitors like ThetaRay, Hawk.ai, and Complidata, all offering solutions that use ML as the primary means of behavior detection and also unsupervised ML to uncover new criminal typologies. The market is maturing as all of these solutions are already in operation within Tier 1 and Tier 2 banks.
Enhancing the Effectiveness of Financial Crime Prevention
The second theme revolved around improving our effectiveness in combating financial crime. This theme was addressed in multiple Anti-Financial Crime panel sessions up on stage as well as in conversations with bankers and vendors. The discussion focused on making financial crime prevention more effective in three significant ways:
-
Increasing Private/Public Collaboration. Given the high number of Suspicious Activity Reports (SARs) filed annually, law enforcement can only follow up on a fraction of these cases. How do we make the most of bank and law enforcement resources to have a material impact on crime reduction. Dan Tannebaum from Oliver Wyman led a session that grappled with this question. The panel included AFC executives from Citi and MUFG as well as an Assitant Secretary from the US Treasury Department and the Head of Policy for the UK's Payment System Regulator (PSR). It was interesting to hear both the banks and the regulators present on what needs to be done. Collaboration between the public and private sectors, especially with a focus on specific types of crimes like human trafficking, was highlighted as essential to significantly reduce the incidence of financial crime.
-
Enhanced Information Sharing. The AI era places a premium on larger datasets. Both banks and regulators recognize that by sharing information, a broader range of illicit activities can be more accurately detected. However, data privacy regulations often hinder such information sharing. While regulators generally support banks in sharing data on financial criminals, there was a call for more explicit guidance on acceptable forms of data sharing, regulators to play a more active role in establishing industry consortia, and the relaxation of data privacy regulations to foster more bilateral collaboration between banks
- Detecting Evolving Criminal Typologies. Rules-based behavior detection systems can only detect known criminal typologies. To stay ahead of emerging threats, banks are now turning to unsupervised machine learning to uncover new criminal typologies. Unlike supervised machine learning, which requires predefined data patterns and labeled data, unsupervised machine learning can identify patterns and relationships in data without knowing what to look for, helping banks to detect new signals and adapt to evolving money laundering techniques. Tier 1 banks are building this capability into their transaction monitoring systems themselves. Smaller banks with smaller data science teams can access this capability through transaction monitoring vendors who built it into their current products.
These key themes will also be featured in our upcoming Risk 2024 advisory, set to be released at the end of this month. Stay tuned for more in-depth insights into the evolving landscape of risk and compliance.