MAS flags risks of using AI agents in digital-asset trading

Chia Der Jiun, managing director of the Monetary Authority of Singapore said innovation must be founded on trust and stability.

SINGAPORE – As artificial intelligence agents increasingly move beyond analysing information to executing transactions, this gives rise to the question of managing risk, especially in the areas of liquidity and collateral, said Chia Der Jiun, managing director of the Monetary Authority of Singapore (MAS), on Oct 9.

“We are entering a world where AI agents may increasingly act on behalf of individuals and institutions, not only to analyse information, but also to initiate and execute financial transactions and other consequential actions,” he noted.

Chia was speaking at a conference organised by business school INSEAD at its Asia Campus. The conference, The Future of Digital Finance: Infrastructure, AI and Regulation, brought together academics, industry leaders and policymakers.

With the foundational capabilities of blockchain and digital assets already proven, the focus has shifted to commercial viability and trust.

“The question is no longer simply whether these technologies work,” he added. “The more important question is what is needed to operate safely, commercially and at scale. Innovation must be founded on trust and stability if it is to scale.”

As the use of AI agents increasingly intersects with tokenised assets and digital money, Chia warned of “significant implications” for the movement of capital.

Because customer agents can autonomously optimise and execute trades, this could result in money moving around faster “for smaller differences in benefits and incentives”, he said, noting that financial institutions will thus have to “sharpen their value proposition”.

He pointed out that this acceleration poses implications for risk management, particularly regarding liquidity and collateral.

“While agents create significant opportunities for efficiency, accessibility and innovation in financial services, they also have to be deployed safely,” he said. “This then requires controls around authority, permissions, identity, accountability and governance.”

Chia added that financial institutions “must be able to ensure” that AI agents act within the scope of a user’s instructions, institutional policies and risk limits, while complying with regulatory requirements.

Earlier in 2026, MAS and industry participants published the Safeguards for Agentic Finance at Runtime (SAFR) white paper. Additionally, through the Future of Finance Institute, MAS has released an open-source reference implementation of SAFR to support industry experimentation.