MAS warns AI agents: Capital migration risks weigh on banks.

CN
1 day ago

On November 19, a single-source report from Bloomberg stated that the Monetary Authority of Singapore (MAS) has rarely pointed its finger directly at AI agents that are transitioning from "analyzing information" to "placing orders": in a world where they can continuously optimize and execute trades 24/7, as long as there are minimal differences in returns or incentives between different institutions, funds can be withdrawn or injected in bulk and synchronously by agents, far exceeding the speed that manual processes can control. Once multiple agents react similarly to the same signal, the collective flow of funds in the same direction can be amplified, potentially rearranging or even depleting the liquidity and collateral originally dispersed across banks, payment institutions, and tokenized deposit networks within minutes, thus pushing financial stability to new uncertain boundaries. MAS publicly called out liquidity management and collateral allocation as the core battleground for these new types of risks, requiring institutions to ensure that agents must operate within the bounds of user instructions, internal policies, and risk limits, and that authorization, permissions, identities, accountability, and governance must be rebuilt. This warning is not an isolated voice but part of a regulatory toolkit along with the previously released "Runtime Agent-Based Financial Protection" (SAFR) white paper and its open-source reference implementation, and the announcement on November 18 by the Future Finance Institute (FFI) to establish an "agent-based currency sandbox": testing agent-initiated payments and treasury management in a controlled environment, embedding risk control into new infrastructures such as tokenized deposits, ultimately bringing crypto financial institutions wishing to take on cross-platform capital flows into a compliance testing ground founded on "controllable capital migration."

From Analytical Assistant to Order Engine: The Concerns of Collective Fund Flows

As AI agents evolve from being "monitoring assistants" to "direct order execution engines," the rhythm of fund movement is no longer determined by human hands and processes but is driven by algorithms. MAS noted in this public warning that such agents can continuously optimize and execute trades, far faster than manual processes: they not only analyze market conditions and assess returns but can automatically convert judgments into orders and payment instructions under the same logic, seamlessly facilitating fund flows between accounts. Once agents start actively reallocating positions and transferring funds across institutions based on preset strategies, the differences in returns and incentives between institutions are no longer a slow game but can be amplified into immediate migration triggers.

What particularly concerns MAS is how funds will collectively act when the insignificant yield or incentive differences that these executing agents focus on are not substantial price discrepancies of a few percentage points but merely a few basis points. In its description, if multiple agents react similarly to the same market signal or risk prompt, the scale and speed of funds flowing towards the "better" end will be magnified; positions that were originally scattered across different banks or platforms could quickly be directed simultaneously by algorithms toward only a few target institutions. Such rapid migration would place direct pressure on the weak links of liquidity management and collateral allocation, and currently, MAS has not disclosed any quantitative thresholds for migration scale or speed, which means institutions must independently anticipate the impact boundaries of such synchronized fund flows without clear red lines.

SAFR Safety Net: Drawing Boundaries for Authorization and Accountability for AI Agents

According to MAS, to control this type of "automatic migration" of funds without quantitative red lines, relying solely on post-monitoring is already too late. Therefore, earlier, the industry jointly proposed the "Runtime Agent-Based Financial Protection" (SAFR) white paper. The positioning of SAFR is not about drafting another set of abstract principles but rather focuses on how to embed risk control directly into the execution chain at the moment "agents are in action" in the context of agent-based finance. The open-source reference implementation subsequently provided by the Future Finance Institute (FFI) serves as a model: any institution preparing to allow AI agents access to real accounts and trading instructions can design its own protective barriers within this framework, rather than merely keeping risk management in the approval process and post-auditing.

In its latest warning, MAS has tightened the boundaries of agent behavior significantly—every action must fall within the intersection of user instructions, internal institutional policies, and established risk limits. This requirement essentially distills SAFR into five control dimensions: who has the authority to allow agents to act (authorization), what agents are allowed to do (permissions), how the system confirms their "identity," who is responsible in the event of an error (accountability), and who keeps modifying these rules (governance). For any bank or platform attempting to let AI agents directly initiate payments, allocate positions, or place orders, these five dimensions are no longer "best practices" but the compliance baseline for accessing the entire financial system.

Agent-Based Currency Sandbox: Testing Tokenized Payments in a Regulatory Environment

After formulating SAFR as "five gates," Singapore's regulatory authorities did not stop at written rules. At the fintech festival on November 18, the Future Finance Institute (FFI) unveiled the "agent-based currency sandbox": in a meticulously defined experimental area, allowing banks and payment institutions to hand over real fund flows to AI agents for initiating payments and cash management, provided that all actions are pre-written into the "fence" of authorization, limits, and risk control. Public information indicates that the sandbox can set key control elements such as the upper limits on transaction amounts per transaction and per day, account access permissions, and real-time risk control rule trigger conditions, but specific parameters and the list of participating institutions have yet to be disclosed, which means it remains in the stage of building regulatory tools and controlled testing rather than fully opening up agent-based payments.

This experimental area did not come into being out of thin air but is superimposed on Singapore's existing tokenized architecture. Under the same BLOOM program framework, StraitsX, OCBC Bank, and Standard Chartered Bank have previously completed the concept verification of interoperability of tokenized deposits, with StraitsX providing settlement infrastructure to interconnect the tokenized deposit systems of the two banks, going through the entire process from initiating payments, compliance screening, liquidity checks, to settlement and confirmation, the core concern being whether compliance filtering and liquidity management can remain controllable and auditable once "deposits are tokenized." Currently, the agent-based currency sandbox has not been explicitly written as an extension of this experiment, but both share a high degree of alignment in regulatory direction: the former expands "who initiates transactions" from people to AI agents, while the latter shifts "which path transactions take" from traditional accounts to tokenized networks, locking the speed and path of fund migration into an observable, limitable framework.

Global AI and Crypto Regulatory Divergence: Singapore Steady, the U.S. Fast

When MAS employs the sandbox and white paper to confine AI agents into an "observable, limitable" framework at the fintech festival, on the other end, political signals are releasing a completely different rhythm. On November 19, while MAS emphasized that agents could amplify synchronized fund flows, impacting liquidity and collateral management, U.S. President Trump simultaneously stated that artificial intelligence is almost entirely a good thing, acknowledging the existence of negative factors, yet explicitly opposing regulatory measures that would "stifle" the industry, even stating that such regulatory occurrences would not be allowed. Singapore chose to first place "machine orders" within a concrete risk control context before discussing scale and scenarios; at the U.S. highest political level, the priority is to endorse innovation itself, viewing constraints on AI as risks to guard against.

However, on the ground of regulatory implementation, the U.S. presents a different fast-paced tech finance landscape. In mid-November, the crypto financial institution Backpack obtained three licenses (DCO, DCM, and FCM) on the 13th, 16th, and 18th, respectively, for engaged regulated derivatives and futures-related business, with the dense timing of license issuance forming a stark contrast to MAS, which is still in the phase of building tools and controlled experiments. The commonality between the two places is their attempts to incorporate fund flows driven by new technologies—whether trades triggered by AI agents or margin and settlement on crypto derivatives—into a licensable and rule-based system, while the differences lie in the path choices: Singapore starts from financial stability, first implementing "runtime protection" and the agent-based currency sandbox, then allows business to trial within protective barriers; the U.S. emphasizes not "stifling" innovation at the top-level discourse while quickly issuing licenses for new crypto financial activities to directly enter the existing regulatory framework, resulting in a global regulatory landscape presenting a bifurcation pattern of tightening runtime risks on one end and accelerating business deployment on the other.

The Next Steps for Platforms and Banks: Seeking Stability Between AI and Tokenization

In MAS's view, the SAFR white paper, open-source implementation, and "agent-based currency sandbox" have quietly pushed the compliance boundary forward: banks, payment institutions, and the tokenized deposit cooperative ecosystem represented by StraitsX, if they were to allow AI agents to initiate transactions, payments, or treasury allocations in the future, would no longer merely view it as an "analytical tool" but must layer in authorization, permissions, identity verification, risk limits, and accountability chains into the system architecture and operational processes before going live. SAFR offers a reusable runtime control framework, while the sandbox provides scenarios for "testing" agent payments and liquidity management under controlled limits and risk boundaries; the payment initiation, compliance screening, liquidity checks, and settlement processes involved in the tokenized deposit interoperability trial also provide references for how to embed agents without breaching protective barriers in the future. For platforms and banks, the next step is not to wait for a set of detailed rules to materialize but to actively view the integration of AI with tokenized operations as the same risk chain to reconstruct internal controls within the existing tools and experimental spaces, as up to now, MAS has not announced any legally binding specialized regulatory rules or timelines, and whether it ultimately upgrades to mandatory regulation, when it occurs, and whether it tightens or loosens largely depends on feedback from sandbox tests and the competitive direction among other major jurisdictions in AI and crypto financial regulation.

Join our community, let’s discuss and become stronger together!
AiCoin exclusive Hyperliquid benefits: https://app.hyperliquid.xyz/join/AICOIN88
AiCoin exclusive Aster benefits: https://www.asterdex.com/zh-CN/referral/9C50e2
On-chain Telegram community: https://t.me/AiCoinWhaleData
On-chain community: https://www.aicoin.com/link/chat?cid=N6OVMor5g
AiCoin on-chain Twitter: https://x.com/aicoinwhaledata

免责声明:本文章仅代表作者个人观点,不代表本平台的立场和观点。本文章仅供信息分享,不构成对任何人的任何投资建议。用户与作者之间的任何争议,与本平台无关。如网页中刊载的文章或图片涉及侵权,请提供相关的权利证明和身份证明发送邮件到support@aicoin.com,本平台相关工作人员将会进行核查。

Share To
APP

X

Telegram

Facebook

Reddit

CopyLink