Three of Asia’s most significant financial regulators are independently moving toward similar expectations for AI governance. They’re arriving from different regulatory traditions, using different vocabulary, and operating on different timelines. But the substance is converging.
MAS published a consultation paper this week that closes in April 2026 — the first APAC prudential regulator to propose board-level AI accountability, mandatory AI inventories, and materiality assessments as supervisory expectations. The framework covers the full lifecycle: data governance, transparency, fairness, third-party AI risk, and human oversight, scaled proportionate to AI complexity. It explicitly covers GenAI and agentic systems.
PBOC’s 2026 Technology Conference established ‘安全有序’ (safe and orderly) as the explicit framing for AI deployment on the mainland. The sandbox approval process now functions as a de facto approved-use-case registry — if a use case hasn’t cleared the sandbox, it’s implicitly out of scope for production deployment. Eighty-five percent of current sandbox batches are AI-themed.
HKMA already has existing circulars and SPM guidance covering AI in banking. The direction of travel is consistent.
What’s notable is the structural similarity beneath the different regulatory languages. All three are converging on: AI inventories (know what you have), materiality assessments (know what matters), lifecycle controls (govern the full pipeline, not just deployment), and proportionate human oversight.
For a bank regulated by all three — and there are several significant ones in this position — the instinct might be to treat these as three separate compliance exercises, each mapped to its respective regulator on its own timeline. That instinct is expensive.
The alternative: identify the superset of requirements across all three frameworks now, while MAS’s consultation is still open. The MAS consultation window is the active lever — submitting a response during consultation earns regulatory goodwill and creates an opportunity to influence the final text. More importantly, building to the superset means a single internal framework satisfies all three rather than three parallel governance silos that drift apart over time.
The convergence is likely not coincidental. Regulators in this region talk to each other. APAC regulatory coordination on AI is an emerging pattern, not three independent processes. Banks that recognise this early and build accordingly will find themselves ahead of the next round of examination questions rather than responding to them.