Jakarta, INTI - The Singapore government has officially introduced the Model AI Governance Framework for Agentic AI, a regulatory framework designed to oversee the development and deployment of autonomous AI systems. The policy is widely seen as a major milestone in ensuring the safety, accountability, and reliability of AI technologies capable of making independent decisions and taking autonomous actions.
The framework specifically highlights the inherent risks associated with Agentic AI, including unauthorized actions, data misuse, and potential systemic disruptions that could have wide-reaching impacts on organizational operations.
Responding to the launch, Keeper Security’s SVP APAC and Japan Country Manager, Takanori Nishiyama, described Singapore’s move as a progressive approach toward building responsible AI oversight.
“This framework emphasizes the importance of securing Agentic AI through an identity-first approach and outcome-based controls. Each AI agent must have a verifiable identity, task-specific permissions, time-bound access, and human accountability at every stage,” Nishiyama stated in a written release.
He further explained that this approach aligns with current enterprise needs, where visibility, control, and auditability are critical factors in large-scale AI implementation. According to him, without a strong foundation of identity management and access control, autonomous AI systems could introduce new security risks.
Zero-Trust as the Foundation for Agentic AI Governance
In its implementation, Nishiyama emphasized that modern Privileged Access Management (PAM) built on zero-trust principles is highly relevant in supporting Agentic AI. This technology removes implicit trust and continuously validates identity, context, and intent, including for non-human entities such as AI agents.
“Continuous monitoring and outcome-based restrictions enable organizations to detect deviations in AI behavior, prevent privilege escalation, and maintain trust in autonomous operations,” he added.
Singapore’s governance framework also underscores key principles such as granular identity management, bounded access, traceability, and auditable decision-making. According to Nishiyama, these principles not only serve as compliance guidelines but also establish a global benchmark for the secure and responsible management of autonomous AI systems.
Furthermore, a lifecycle-based approach covering development, testing, deployment, and ongoing monitoring is considered essential to ensuring security in environments where AI operates at machine speed.
“By embedding security from the outset, organizations can fully leverage AI capabilities while maintaining trust, control, and regulatory compliance,” Nishiyama concluded.
Conclusion
Singapore’s governance framework for Agentic AI demonstrates how zero-trust security, identity-based controls, and continuous oversight can form a strong foundation for responsible autonomous AI adoption. By aligning technological innovation with robust governance principles, organizations can harness AI’s transformative power while minimizing emerging security risks.
Read more: Barnsley Repositions Itself as the UK’s First “Tech Town” Amid Growing AI Investment from US Tech Giants