Singapore Launched the World's First Governance Framework Specific to Agentic AI

While most AI regulatory frameworks are still thinking about chatbots that answer questions, Singapore is already writing the rules for agents that act on their own.

The world's first dedicated to agentic AI

Singapore unveiled its new Model AI Governance Framework on January 22, 2026, at the World Economic Forum — by its own account, the world's first to provide guidance dedicated specifically to managing risk in the deployment of agentic AI. It's a direct evolution of the original 2019 Model AI Governance Framework, now extended to a type of system that barely existed commercially at that date.

What makes agentic AI different, according to the framework

The document defines AI agents as systems that can plan, reason, and take autonomous actions independently to achieve goals on behalf of a user — initiating tasks, updating databases, executing actions, and adapting dynamically. It identifies specific risks from this capability: unauthorized actions, data leakage, and biased decisions made without direct human oversight in the moment.

How it connects to AI Verify

The framework doesn't replace Singapore's existing tools — it complements them. AI Verify, the government-developed AI testing toolkit, validates AI system performance against internationally recognized principles, and now conceptually extends to also cover validating autonomous agent behavior, within the same ecosystem that includes the Global AI Assurance Pilot launched in 2025.

Why Singapore usually goes first

Singapore has a consistent pattern: it's voluntary (not a law with sanctions), but it's widely adopted across the APAC region as a de facto reference, and it frequently anticipates by 12-24 months the topics other regulators are just beginning to discuss. That makes it a useful early signal — the risks Singapore identifies today in a voluntary framework are, with good probability, the same ones that will show up in mandatory laws in other jurisdictions within a year or two.

What it means for companies building autonomous agents

For any company deploying AI agents capable of executing real actions (not just answering questions) — the kind of architecture we covered in our Strands Agents SDK guide — Singapore's framework is, in practice, the first risk checklist specifically designed for that use case, useful even for companies that don't operate in Singapore and simply use it as a best-practices reference.