Use of Artificial Intelligence in Supervision of Financial Institutions

Mr Desmond Choo asked the Prime Minister (a) how has the Monetary Authority of Singapore used artificial intelligence (AI) in its supervision of financial institutions; (b) whether this has resulted in positive outcomes; and (c) what is the long-term role of AI in its regulatory framework.

Financial supervision is primarily a data-driven activity. The Monetary Authority of Singapore (MAS) adopts a risk-based approach to its supervision, which entails focusing more supervisory resources on riskier financial institutions (FIs), as identified by our data, frameworks and processes.

The advancement of data analytics techniques like artificial intelligence and machine learning (AIML) has expanded the toolkit for MAS to make better sense of the various risk signals in the voluminous data it receives. This has allowed us to automate certain tasks that used to require manual processing, as well as better identify outliers or suspicious networks for closer scrutiny.

I will illustrate with two broad areas that have yielded meaningful results for us.

First, MAS has developed tools using machine learning to improve our risk targeting for supervisory or enforcement action. For example, MAS has trained a machine learning model, using traits identified by human experts, to analyse market trading data to help enforcement officers identify and prioritise potential market collusion or manipulation for investigation. MAS also employs machine learning to help supervisors identify financial advisory representatives who may present higher risks of exhibiting bad behaviours, such as mis-selling investment or insurance products. FIs with a greater number of representatives presenting higher risks of engaging in mis-selling will be prioritised for deeper supervisory engagement.

Second, MAS also applies natural language processing (NLP) to help supervisors work more efficiently. Instead of having supervisors manually trawl through voluminous textual data, such as reports submitted by FIs, MAS uses NLP to analyse the texts and flag issues for supervisors’ attention. MAS also uses NLP to scan social media and industry or media analysis reports for news and developments that may warrant supervisory attention.

Apart from AIML, MAS also uses advanced data analytics to identify networks of suspicious activity in our financial system, which may indicate money laundering, terrorism financing or other financial crimes. MAS then engages FIs to warn them of potential threats and to assess the robustness of their controls. Such data analysis techniques combined with more powerful machine learning tools can help MAS sift out high-risk networks and transaction patterns more effectively.

MAS will continue to explore how latest technologies, including AIML and generative artificial intelligence solutions, can be deployed responsibly and securely to enhance financial supervision.