## A New Framework for AI Agents in Singapore

Singapore has launched an important new Model AI Governance Framework (MGF) for Agentic AI, announced on January 22, 2026, by the Ministry for Digital Development and Information. This framework was created by the Infocomm Media Development Authority (IMDA) and is designed to help companies safely use AI agents in their business.

AI agents are special computer programs that can plan and make decisions on their own to reach goals. They are like robots that can think about problems and figure out what to do next without someone telling them every step. The Singapore framework helps companies understand the risks that come with these smart programs and how to manage them responsibly.

## Understanding the Challenges of AI Agents

AI agents bring new challenges that regular AI systems don't have. These agents often have the power to make choices, work with other agents, and even change their own approach as they learn. Because of this independence and decision-making ability, new types of mistakes can happen.

Experts have discovered that AI agents can create problems that older computer systems never had before. For example, when multiple AI agents work together, they can sometimes make decisions that nobody expected. Traditional computer systems follow a set of rules that never change, but AI agents can adjust what they do based on what happens around them.

## The Global Regulation Picture

Different parts of the world are handling AI regulation in very different ways. In 2025, global AI regulation entered a strong enforcement phase, meaning that rules written in the past are now being put into action. In the United States, the new government has decided to focus less on regulation and more on letting companies move fast with new technology. This is a significant change from the previous approach.

Meanwhile, the European Union is taking a stricter path with its AI Act. Starting in June 2025, the EU required companies to sort their AI systems by how risky they are, prepare plans for watching them, run special tests, and share information about them publicly. AI systems that are used in important decisions—like hiring workers, giving loans, judging school applications, or running government services—face even stricter rules and must be watched constantly.

## What Existing Frameworks Can and Cannot Do

Three main frameworks help companies manage AI systems today: ISO/IEC 42001, the NIST AI Risk Management Framework, and the EU AI Act. Each one has good points, but none of them completely solve the problems that come with AI agents.

ISO/IEC 42001 is good at helping companies organize their AI work. It shows companies how to write down what they're doing and keep improving their systems. However, this framework doesn't explain how to set limits on what an AI agent can do on its own or how to decide who should approve certain choices.

The EU AI Act is the most complete regulation created so far. It requires companies to check their systems closely and get humans to review important decisions. But the EU Act assumes that AI systems will work the same way every time, which is not true for AI agents that learn and change.

## New Solutions: Policy as Code

Companies are now using an exciting new approach called policy as code to control AI agents better. This means taking the rules that a company wants to follow and turning them into computer language that AI agents automatically follow.

Think of it like this: instead of telling an AI agent "be careful with customer information," companies write actual computer rules that stop the AI agent from doing anything with customer information unless it gets special permission. This approach removes human mistakes from the picture and makes sure all AI agents follow the same rules every time.

When companies use policy as code, every decision the AI agent makes gets recorded automatically. This creates a clear trail that companies can look at later to make sure everything happened correctly. For jobs like banking and manufacturing, this is very important because mistakes can cost a lot of money or break laws.

## What Companies Need to Do Right Now

Organizations are facing three big pressures in 2026. First, more and more AI agents are being created and used in companies. Second, these AI agents often have special access to important information, just like a trusted worker would. Third, governments are watching more closely to make sure companies are using AI safely and fairly.

Companies need to keep track of all their AI systems and understand what information each one can see and use. This is challenging because sometimes AI agents have access to sensitive data without anyone realizing it. By watching which AI agents have access to what information, companies can find and fix problems before they become serious.

## The Path Forward

The world is moving toward stricter and clearer rules for AI agents. Some experts think that by 2026, the only acceptable uses of AI will be those that can be explained and justified to regulators. Companies that succeed will be the ones that build safety and responsibility into their AI systems from the beginning, not as an afterthought. As AI agents become more common in businesses, having strong governance frameworks in place is no longer optional—it is essential for protecting companies, workers, and customers.