Blog
Singapore's agentic AI governance framework: Why identity security is the foundation of agentic AI accountability
Singapore's Model AI Governance Framework for Agentic AI is a first-of-its-kind governance framework built specifically for AI agents,1 and identity runs through nearly every part of it. The framework recommends that every AI agent be a known, individually identifiable actor rather than a shared service account, with bounded permissions and a traceable audit trail, as part of how organizations bound risk upfront and keep humans meaningfully accountable.
That's the headline finding for any enterprise operating in Singapore right now. Governance readiness for agentic AI doesn't start with a policy document. It starts with knowing exactly which agents exist, what they can touch, and who's accountable for them. Singapore calls this guidance voluntary. In practice, it's becoming the price of doing business.
Singapore's voluntary AI guidance can still carry practical weight
Singapore has no single AI law. Instead, the Infocomm Media Development Authority (IMDA) has spent years building a layered set of principles-based frameworks, the Model AI Governance Framework, its generative AI extension, and now the Model AI Governance Framework for Agentic AI, launched on 22 January 2026 at the World Economic Forum and updated on 20 May 2026 with real-world enterprise case studies covering multi-agent systems, third-party agents, and automation bias.2
None of this is legally binding. But three factors may give it practical weight:
- Supervisory pressure. The Monetary Authority of Singapore (MAS) references national AI governance norms when setting expectations for regulated financial institutions.
- Regional influence. Singapore's frameworks have a track record of becoming ASEAN and international reference standards, raising the stakes for any regional headquarters.
- Procurement leverage. Enterprise buyers are increasingly writing agentic AI governance clauses into vendor assessments, meaning your governance posture may affect your ability to close deals, not just pass audits.
The framework itself rests on four pillars: assessing and bounding risk, meaningful human accountability, technical controls, and end-user responsibility. Identity threads through the first two especially: the framework recommends that agents need unique, verifiable identities tied to accountable human owners, because current authentication systems were built around a single human per account and don't hold up once agents start acting for multiple users or spinning up sub-agents of their own.
The real gap: Identity management hasn't caught up to non-human identities
Here's the uncomfortable truth buried in the framework's own text: current authorization systems weren't built for agents. They assume static, pre-defined scopes and a single human tied to a single account. Agents need permissions that shift dynamically based on context and risk. They act on behalf of multiple users. They spin up sub-agents that spin up more sub-agents.
Most organizations have mature identity and access management for their human workforce. Few have extended that same rigor to non-human actors, and that gap is exactly where regulatory and security risk can concentrate.
This isn't a Singapore-only problem. Across the 100 security and identity leaders at Fortune 500 and Global 2000 companies SailPoint® surveyed for its own research, 91% of organizations are already deploying or evaluating AI agents, yet only 21% report being "very" or "extremely" confident in their ability to manage the security risk that comes with them.3 Confidence hasn't caught up with adoption, in Singapore or anywhere else.
Why it matters now: Agents that lack unique, cryptographically verifiable identities are difficult to trace, hold accountable, or audit. That's not a theoretical gap. It's the specific failure mode Singapore's framework is designed to address.
What the framework actually recommends for agent identity
According to IMDA's guidance, every agent identity should be:
- Unique and verifiable, the agent can identify itself to the organization, its human owner, and other agents.
- Accounted for, tied to a supervising agent, human user, or department for accountability.
- Differentiated by capacity, recorded according to whether it acts independently or on a human's behalf.
- Catalogued centrally, tracked in a single system to prevent agent sprawl and enable anomaly detection.
Authorization, meanwhile, should be scoped, time-bound, non-transferable, and never exceed what the authorizing human is permitted to do themselves.
That's a precise, actionable checklist. It's also, functionally, an identity governance program.
Why point solutions may not close the agent identity gap
Faced with this, some organizations reach for narrow fixes: a discovery tool here, a monitoring dashboard there. The problem is that AI agents don't always sit still long enough for a piecemeal approach to work. They operate across agentic platforms, browsers, endpoints, and SaaS applications simultaneously, and they act at machine speed, far faster than a quarterly access review or a manually maintained spreadsheet can typically track.
This is the identity-centric problem SailPoint Agentic Fabric™ was built to help solve. Rather than bolting on a separate tool for agent visibility, Agentic Fabric extends SailPoint's enterprise-grade identity security platform to treat AI agents, bots, and machine accounts as first-class governed identities, alongside every human one.
That matters directly against the framework's four pillars:
- Assess and bound risk: Discovery & Unified Visibility is designed to find and register agents and non-human identities into a single identity graph, mapping relationships to human owners, entitlements, and sensitive data, helping address the visibility gap the framework highlights.
- Meaningful human accountability: Agent Identity Security helps establish clear ownership from creation to retirement, so "who is accountable for this agent" has a documented answer, not a guess.
- Technical controls: Real-time Governance & Audit helps enforce least privilege and automated provisioning/deprovisioning with an audit trail, helping close the multi-month gaps left by periodic access reviews with continuous governance.
- Ongoing monitoring: Proactive Protection & Response calculates a dynamic risk score for each agent and can trigger automated remediation, revoking entitlements or disabling agents when behavior deviates from expected patterns, because a human approver can't typically react at the speed an agent operates.
Data Access Governance adds a further layer: mapping the human-to-agent-to-data access path to help prevent agents from bypassing existing controls to reach sensitive information.
Building comprehensive agent identity governance
Organizations navigating Singapore's agentic AI landscape, and the MAS, PDPC, and CSA guidance and requirements that may apply alongside it, should be able to answer these questions with evidence, not assumptions:
- Can we produce a complete, current inventory of every AI agent and machine identity in the enterprise?
- Does every agent have a unique, traceable identity tied to a human owner or department?
- Are permissions scoped to least privilege, with automated deprovisioning when an agent is retired or reassigned?
- Can we detect and respond to anomalous agent behavior in real time, not at the next audit cycle?
- Is there a continuous, immutable audit trail covering every agent action?
If any answer is "not yet," that's not a future problem. It's a present one and it's the kind of gap Main Line Health addressed when it achieved 75% faster access provisioning and deprovisioning, and Seminole State College addressed on its way to a 10% reduction in help desk call volume. Governing non-human identity with that same discipline is the next logical step.
Get Ahead of Agentic AI Risk
Singapore's agentic AI framework may offer a preview of where other regulators are heading: identity as a control point for AI accountability. Waiting for binding rules to catch up isn't a proactive strategy.
Download our The year of AI Agents report to see how identity leaders are governing AI agents today, and what separates organizations that are ready for the agentic era from those still finding out the hard way.
DISCLAIMER: The information contained herein is for informational purposes only, and nothing conveyed herein is intended to constitute any form of legal advice. SailPoint cannot give such advice and recommends that you contact legal counsel regarding applicable legal issues. The development, release, and timing of any features, functionality, capabilities, or services described for SailPoint's products that are not currently available remain at SailPoint's sole discretion on a when, and if, available basis and may not be delivered at all and must not be relied on in making a purchasing or investing decision.
Sources
- Singapore Ministry of Digital Development and Information, “Singapore Launches New Model AI Governance Framework for Agentic AI,” 22 January 2026: mddi.gov.sg.
- SailPoint, “The year of the AI agent” (2026), survey of 100 security and identity leaders at Fortune 500 and Global 2000 companies, polled 22–28 April 2026 (pp. 3, 8–9)