As AI agents move from experimental tools to operational systems with real-world impact, Singapore’s newly launched Model AI Governance Framework for Agentic AI is set to reshape how businesses deploy, manage, and scale these technologies.
Unveiled at the World Economic Forum in Davos, the framework is the first in the world to offer structured, practical guidance specifically for agentic AI, systems capable of planning across multiple steps and taking actions on behalf of users. While not a law, the framework is likely to influence business practices quickly, especially in regulated and customer-facing sectors.
For companies in Singapore, the message is clear: AI agents can drive productivity and transformation, but only if governance is designed into systems from the start.
Unlike traditional or generative AI, AI agents can initiate transactions, update databases or trigger workflows autonomously. This expanded capability raises new risks, including unauthorised actions, data misuse and over-reliance on automated decisions. The framework responds by emphasising that humans remain ultimately accountable, even as autonomy increases.
“Agentic AI systems will make decisions with real-world consequences,” said Elsie Tan, country manager for Worldwide Public Sector, Singapore, at Amazon Web Services, in a press statement issued by IMDA. “We need concrete mechanisms for visibility, containment, and alignment built into infrastructure, along with human judgment to use them wisely. Singapore’s Model AI Governance Framework is a step in the right direction.”
Also Read: Voice does not expire: How AI helps us keep our stories alive
In practical terms, businesses are expected to rethink how AI agents are authorised, monitored and approved. One of the framework’s core recommendations is to assess and bound risks upfront by selecting appropriate use cases and limiting an agent’s autonomy, access to tools and exposure to sensitive data. For enterprises, this means more formal approval processes for agent deployments, especially for systems that can trigger payments, modify records or interact directly with customers.
The framework also elevates the importance of human checkpoints. As AI agents become more reliable, organisations risk automation bias, the tendency to over-trust systems that have performed well in the past. By requiring defined moments where human approval is mandatory, companies can reduce the risk of silent failures or cascading errors.
For tech vendors and cloud providers, the framework may shape how products are built and sold. It encourages technical controls such as baseline testing, lifecycle monitoring and restricted access to whitelisted services, alongside non-technical measures such as training and transparency. These expectations could increasingly become standard requirements in enterprise procurement.
“Building trust in agentic AI is an ongoing, shared responsibility, and IMDA’s framework is a constructive first step,” said Serene Sia, country director for Malaysia and Singapore at Google Cloud.
She added that open standards will play a key role in enabling secure multi-agent systems. “Having pioneered open standards like the Agent2Agent Protocol and Agent Payments Protocol, Google has been playing a key role in establishing the foundation for interoperable and secure multi-agent systems.”
Also Read: Forward-looking governance: Why Asian boards must think like futurists
The impact will be felt most strongly in sectors where AI agents operate close to money, data or safety. Financial services firms, fintech companies and banks are likely to introduce stricter approval gates, audit trails and monitoring to meet expectations of accountability. E-commerce platforms and logistics providers may need tighter controls around customer service agents who can issue refunds or amend orders.
For organisations already deploying AI agents at scale, the framework offers validation and direction.
“At KBTG, we have already begun deploying AI agents across the bank and have a strong pipeline of additional agents ahead,” said Dr. Komes Chandavimol, principal AI evangelist at KASIKORN Business-Technology Group, the technology arm of KASIKORNBANK. “As we move toward deployment at scale, we are strengthening our agentic AI governance. The Model Governance Framework for Agentic AI is a timely and practical document that will help guide this journey.”
Small and medium-sized enterprises may face capability gaps, particularly around testing and monitoring. This could accelerate demand for managed services and “governed-by-design” AI agents that embed compliance features by default.
Positioned as a living document, the framework is likely to evolve alongside the technology. For businesses in Singapore, it sets a clear direction of travel: AI agents are welcome — but only with accountability, oversight and trust built in.
—
The lead image of this article is generated by AI.
The post Singapore’s new AI governance framework signals a turning point for businesses using AI Agents appeared first on e27.