What does MAS Circular MAS/TCRS/2025/06 say about AI in banking?

What are the key points of MAS Circular MAS/TCRS/2025/06?

• Section 2 lists five risk types from deepfakes: market, cyber, fraud, regulatory (for example falsified identities in recruitment by actors from sanctioned countries) and reputational.
• Section 2.1 addresses defeating biometric authentication, with recent incidents and possible mitigating measures.
• Section 2.2 addresses social engineering and impersonation scams, with incidents and mitigation.
• Section 2.3 addresses misinformation and disinformation, including fabricated executive statements and market-moving content.
• Section 3 concludes that FIs should assess deepfake risks to their own operations and customer interactions and keep monitoring both generation and detection technologies.
• The conclusion names liveness detection, additional verification for high-risk transactions, industry information sharing, awareness campaigns, real-time deepfake detection in communication channels and endpoint devices, and regularly updated and tested incident response plans.
• Appendix A summarises threats, impact and mitigating measures; the paper credits the Association of Banks in Singapore Standing Committee on Cyber Security and MAS's Cyber and Technology Resilience Experts panel for input.

What did MAS Circular MAS/TCRS/2025/06 change for banks?

It expands the deepfake material in MAS's July 2024 paper into a standalone treatment aimed at the threats most relevant to banks: biometric onboarding and authentication, impersonation of executives and customers, and disinformation affecting markets and reputation.

What does MAS/TCRS/2025/06 expect banks to do about deepfake risks?

MAS information paper MAS/TCRS/2025/06, 'Cyber Risks Associated with Deepfakes', of September 2025 says financial institutions should assess the risks deepfakes pose to their own operations and customer interactions and implement appropriate defences. It identifies three areas of impact (defeating biometric authentication, social engineering and impersonation scams, and misinformation and disinformation) across five risk types (market, cyber, fraud, regulatory and reputational). Its conclusion names liveness detection, additional verification for high-risk transactions, information sharing, awareness campaigns, real-time deepfake detection integrated into communication channels and endpoint devices, and incident response plans that are regularly updated and tested. It is guidance rather than a binding rule.

The paper complements MAS's July 2024 generative AI cyber paper and is cited in MAS's November 2025 consultation paper as the paper covering AI used against FIs by third parties, which the proposed AI risk management Guidelines exclude from their scope.

For banks the practical overlap is with onboarding and know-your-customer controls, payment authorisation and fraud operations, since the same deepfake techniques are used to defeat biometric checks and to impersonate customers or executives.

WHAT THIS MEANS IN PRACTICE

• Test onboarding and authentication flows against deepfake and injection attacks.
• Verify unusual or urgent payment and information requests through a separate, trusted channel (the paper gives the example of confirming a video call by another medium).
• Include deepfake impersonation of executives in tabletop exercises.
• Join industry information-sharing channels on deepfake incidents.
• Review the Appendix A threat and mitigation table against current controls.