A significant debate has been playing out between governments and the AI makers over who controls the boundaries of how artificial intelligence gets used. Governments want to deploy AI at whatever scale and in whichever way they see fit, without restrictions. AI makers are increasingly wary of providing that kind of unfiltered access. The tension between those two positions is not just a contractual dispute- it is a signal of a much wider governance problem that every organisation operating in or around the public sector needs to take seriously.
The stakes are particularly acute across Asia Pacific. According to Forrester's 2025 IAM Outlook for APAC, machine identities have become the region's fastest-growing and most vulnerable attack surface, accelerated by the rapid adoption of generative and agentic AI. SailPoint research reinforces this: 82 percent of organisations deploying AI agents still lack clear accountability for what those agents access, what decisions they influence, and who is responsible when something goes wrong.
At the heart of this is an identity problem. Every AI agent operates with real permissions, accesses real data, and makes real decisions. Without knowing who and what has access to what, and under what authority, governance is not possible. For enterprises across banking, telecoms, healthcare, and critical infrastructure, that identity gap carries consequences that extend well beyond internal security.
When a government deploys AI without adequate guardrails, the exposure does not stay contained. Every technology vendor, contractor, and service provider connected to that government inherits a share of the risk, whether they have visibility into it or not.
For a regional bank processing government payments, a telco managing public sector infrastructure, or a healthcare provider operating under government contracts, ungoverned AI at the top of the supply chain creates vulnerabilities at every point below it.
The exposure is not limited to operational systems either- consumer data, citizen records, and personal information processed by AI agents without proper identity controls become liabilities shared across the entire ecosystem.
Across Asia, regulators are already moving. Indonesia's Personal Data Protection Law now requires businesses to localise sensitive data and strengthen cross-border protections. In Vietnam, the AI Law took effect in March 2026, establishing a comprehensive legal framework covering developers, providers, and users of AI systems.
Singapore has gone further still, launching the world’s first Model AI Governance Framework for Agentic AI, offering structured guidance on responsible agentic AI deployment, from bounding agent autonomy upfront to maintaining meaningful human oversight.
The direction is clear. But regulation that arrives after AI systems are already deeply embedded is harder to apply and more disruptive to enforce than governance built in from the start. For governments and the enterprises connected to them, three things cannot be deferred:
- Define boundaries and assign ownership before deployment
Governance frameworks designed after deployment are not governance frameworks; they are damage mitigation. AI agents embedded in workflows accumulate access, develop dependencies, and integrate into operational infrastructure in ways that make retroactive constraints difficult to enforce.
The organisations that consistently manage AI risk effectively define operational parameters upfront: what a system is authorised to do, what data it can access, what falls within its autonomous authority, and what requires human approval without exception.
This includes user data. AI agents that can access, process, or share personal information without defined boundaries and clear oversight are not just a security risk. They are a regulatory one, particularly across APAC where data protection obligations are tightening simultaneously across multiple jurisdictions.
Equally critical is the question of ownership. Systems get deployed, integrated into critical processes, and then the people responsible for them move on. Access parameters go unreviewed and behaviour goes unmonitored. In government contexts, diffused accountability is a public trust failure. In enterprise contexts, particularly for banks, insurers, and healthcare providers, it is a material compliance risk.
Every deployed AI system requires a named, accountable owner at all times, with formal protocols for continuity when personnel change. What is required is an adaptive approach to identity governance, one that continuously evaluates and adjusts access and permissions as AI systems evolve, not just at the point of deployment but throughout the life of the system.
Boards and executive teams should be asking a simple question: for every AI system operating in this organisation, can we name the person accountable for it today?
- Eric Kong, Global Vice President, ASEAN, SailPoint.
- Make human oversight structurally enforceable
AI systems, however capable, lack the contextual judgement that consequential decisions require. They cannot weigh the public interest implications of a procurement decision, assess the ethical dimensions of a citizen-facing outcome, or recognise when an edge case falls outside the scope of their original mandate. These are inherently human responsibilities, and no degree of technical sophistication changes that.
The question is not whether humans should be in the loop. It is whether the systems and processes around AI genuinely empower humans to intervene, or whether human oversight is simply stated on paper while AI operates without meaningful accountability in practice. Effective oversight is continuous, contextual, and structurally enforced, not an afterthought applied when something goes wrong.
- Build a two-way relationship with government on governance
For enterprises connected to the public sector, the temptation is to treat AI governance as an internal matter and wait for governments to set the regulatory terms. That calculation is becoming harder to sustain. As governments across Asia scrutinise their technology supply chains more carefully, governance maturity is increasingly a condition of participation, not just a compliance requirement.
Organisations that can demonstrate clear accountability for their AI systems, defined access controls, and auditable oversight are better placed to maintain and grow public sector relationships.
But the opportunity goes further than compliance. Enterprises that engage proactively with governments on AI governance, contributing to the development of frameworks, sharing operational insights from deployment, and helping to define what responsible AI looks like in practice, become genuine strategic partners rather than vendors to be scrutinised.
In a region where AI governance frameworks are still being shaped, that positioning carries long-term value that reactive compliance never will.
What enterprises should do now
The regulatory window is narrowing and supply chain scrutiny is intensifying. Enterprises across APAC, particularly those operating in or adjacent to government supply chains, should take three immediate steps.
First, conduct a full inventory of every AI agent operating across the organisation, including third-party tools embedded in workflows, and assign a named owner to each. Without visibility, governance is not possible.
Second, review the access permissions of existing AI systems against the principle of least privilege. Agents should have the minimum access required to perform their function. Permissions that were granted at deployment and never reviewed are one of the most common sources of uncontrolled exposure.
Third, bring AI accountability into board-level risk oversight. The question is not just whether AI is being used, but whether the governance architecture around it would withstand regulatory scrutiny today. In an environment where governments are tightening supply chain requirements across the region, that is no longer a question that can sit below the board.
The enterprises that get this right will not just avoid risk. They will be better positioned to grow, to win public sector relationships, and to deploy AI with the confidence that comes from knowing the foundations are sound.
Eric Kong is the Global Vice President for ASEAN region at SailPoint.
