Generative artificial intelligence, or Gen AI, is an increasingly powerful tool that generates text, images, and videos. While the most well-known use of Gen AI is ChatGPT, Google’s recently launched AI video generator Veo 3 went viral for its ability to generate audio that syncs seamlessly with the video. Unfortunately, this also means it is easier than ever for malicious actors to exploit this technology for criminal gain via scams, fraud, and market manipulation.
The Singapore government recognises that Gen AI is a rapidly evolving technology that will transform the financial sector by improving efficiencies and personalising customer experiences. However, it is also wary of the risks that the same technology will bring, not just for crime but also data risk, and introducing unintended biases into the financial sector.
Read Also: 4 Trends We Learned At The Singapore FinTech Festival (SFF) 2024
This is why the Monetary Authority of Singapore (MAS) launched an initiative called Project MindForge, which examines the risks and opportunities of Gen AI for financial services. The goal is to develop a clear and concise framework for the responsible use of Gen AI within the financial industry. This project is supported by a consortium including local and foreign banks, the Association of Banks in Singapore, Google Cloud, Microsoft, and Accenture.
The framework consists of seven risk dimensions for Gen AI especially in the finance industry.
#1 Fairness And Bias
Since Gen AI relies heavily on data input, it is important to set fairness objectives to help identify and address unintentional bias and discrimination. Poorly sourced data from the Internet could have adverse or inappropriate impact on individuals and groups.
Mitigation measures in the finance industry include extensive testing by industry experts and built-in safeguards to reduce the risks.
#2 Ethics And Impact
One notable danger when using Gen AI is the inadvertent appearance of toxic and offensive output, or more insidiously, unintended deceit and manipulation of users. This is often due to a lack of clearly defined core values and ethics that must be responsibly programmed into the AI.
Mitigation measures in the finance industry include responsible AI guardrails such as toxicity checks, which help in reducing false or misleading outputs generated by AI, also known as hallucinations.
#3 Accountability And Governance
To complement the earlier two risk dimensions, “fairness and bias” and “ethics and impact”, this risk dimension ensures adequate human oversight to take accountability for the outcomes and impact of the use of AI. Should there be a reliance on third parties, accountability must be clear and enforceable.
Mitigation measures in the finance industry include the need for specific guidance by the appropriate subject matter experts. For example, if AI is used in Anti-Money Laundering and credit risk assessment, then the appropriate compliance industry experts should be consulted.
Read Also: GFTN: How Singapore’s Global FinTech Ambitions Rest On The Shoulders Of This Newly-Minted Business
#4 Transparency And Explainability
Since Gen AI results are interpreted and audited by human users, the origins of the AI’s training and test data must be clear and transparent. In addition, adequate feedback and recourse mechanisms must be introduced to ensure the AI’s accuracy.
Mitigation measures in the finance industry include multiple rounds of human testing, using a specialised testing user interface to collect feedback to ensure that the AI tool undergoes sufficient refinement till it achieves a clearly defined accuracy rate.
#5 Legal And Regulatory
This is a crucial risk dimension for AI to be used reliably, especially in the highly regulated finance industry. It is important to avoid unauthorised data transfer and storage, especially of private information, and to ensure sufficient intellectual property (IP) protection to avoid any IP infringement.
Mitigation measures in the finance industry would once again involve the guidance of subject matter experts with the necessary domain knowledge to ensure no breaches or misalignments with regulatory or organisational standards.
#6 Monitoring And Stability
The systems behind Gen AI must be consistently maintained to ensure that the output it generates is not misinterpreted. This includes ensuring a robust system for monitoring and managing AI applications.
Mitigation measures in the finance industry include incorporating relevant external knowledge into the generation process to produce responses that are based on factual information and contextual relevance.
#7 Cyber And Data Security
Finally, AI systems must be protected from cyberattacks, unauthorised access, data loss, and misuse by malicious actors.
Mitigation measures in the finance industry should ensure that the AI systems in use have the same high level of protection as other internal systems used by financial institutions.
The next phase of Project MindForge will expand the framework to involve financial institutions from the insurance and asset management industries. It also aims to develop an AI governance handbook for the financial industry. While there is no timeline for the end of the next phase, an upcoming panel at the SuperAI event in June may give us an update.
Mr. Rajeev Hassamal, DBS Singapore’s Head of GenAI Future of Work, and Mr. Kenneth Gay, Chief FinTech Officer, MAS will be on a panel entitled “New Models, New Machines: AI in Finance Today” at 11.15am on June 19th 2025. The panel will be moderated by Mr. Jaskaran Bhalla, Head of Content, Global Financial & Technology Network, a non-profit organisation launched by MAS.
To sign up, head to the SuperAI website for more information on Asia’s largest AI event, held from 18 to 19 June at Marina Bay Sands.
