Raymond Cheung
Chartered Actuary · CRO · Board Adviser · Singapore
MAS's Managing Director used his 11 September 2026 Global FinTech Fest address to put SAFR — the industry framework for governing autonomous AI agents in finance — back in front of the market, alongside a sixfold rise in high-severity vulnerabilities this year. For boards outside the banks and fintechs that co-authored it, that repetition is the signal worth reading, not the framework itself.
A board I've advised spent an hour last quarter satisfied that its AI governance was in order because it had a policy on chatbot use. That is not the conversation MAS is having anymore. On 11 September, Managing Director Chia Der Jiun told the Global FinTech Fest that AI would increasingly strip out transaction and decision-making frictions altogether — and in the same address, flagged that high-severity vulnerabilities tracked this year had risen sixfold to around 2,200. Read together, that is a regulator describing autonomous systems taking actions on their own, faster than the institutions running them can watch every one, at the exact moment the attack surface is expanding fastest.
The white paper MAS keeps putting back in front of the industry
MAS published SAFR — Safeguards for Agentic Finance at Runtime — as an industry white paper in July 2026, developed under its BuildFin.ai initiative with Ant International, Circle, HSBC, J.P. Morgan Chase, Manulife, Mastercard, OCBC and Visa. It is explicitly not regulatory guidance; MAS says so in the document itself. But a Managing Director does not use a keynote nine weeks later to walk an industry audience back through the same territory — AI agents, oversight, runtime controls — unless the regulator intends it to become the reference point institutions get measured against before it becomes a Notice.
“SAFR isn't a rule yet. MAS repeating it in a Managing Director's keynote nine weeks after publishing it is not an accident.”
The four questions it forces a board to answer
Strip away the terminology and SAFR asks a board four plain questions about every AI agent operating anywhere near money movement, underwriting, or claims decisions: does the agent have a verifiable identity distinct from the human who deployed it; is there a controls repository defining what it is and isn't permitted to do; is there a disposition engine that evaluates each proposed action and routes it to Deny, Escalate, Auto-Execute or Observe; and is there an audit log detailed enough to reconstruct, after the fact, exactly why the agent did what it did. Most boards outside the SAFR co-authors cannot currently answer any of the four for the AI tools already running inside their operations — not because the technology is unusually complex, but because the tools arrived through procurement or an ops team's pilot project, not through a governance conversation.
- Can management name every AI agent currently authorised to take an action — not just answer a query — inside the business, and who approved that authorisation?
- For each one, is there a defined trigger that escalates the decision to a human, or does the agent simply act and report afterward?
- If a regulator or auditor asked for the audit trail behind a specific automated decision from six months ago, could it be produced today?
None of this requires a board to slow down AI adoption — MAS's own message is the opposite, that innovation should scale, not stall. What it requires is a board that treats agent identity, escalation triggers, and audit trails as standing governance questions now, while building the infrastructure is a design choice, rather than in eighteen months when it is a supervisory finding.
Common Questions
Is MAS's SAFR framework a mandatory requirement for Singapore financial institutions?
No. SAFR is an industry white paper published by MAS under its BuildFin.ai initiative with major banks, insurers, and payment firms, and MAS states explicitly that it does not constitute regulatory guidance or supervisory expectations. It is a voluntary reference architecture — but MAS's continued public promotion of it, most recently in the Managing Director's September 2026 keynote, signals the direction future guidance is likely to take.
Which board committee should oversee AI agent deployment in a Singapore financial institution?
There is no prescribed structure, but the practical answer is whichever committee already owns technology and operational risk — typically the risk committee or a technology risk subcommittee — should extend its mandate explicitly to autonomous AI agents rather than leaving oversight to management's AI governance policy alone. The board's role is to confirm agent identity, escalation, and audit-trail controls exist, not to approve individual use cases.
About the author
Raymond Cheung is a Chartered Actuary, C-suite executive and board adviser with more than 20 years of experience across Asia in risk management, insurance, ESG and corporate governance. He is the CEO of CER Consultancy and an accredited trainer at SMU Academy and the Singapore College of Insurance.