Skip to main content
Cybersecurity

Singapore Leads in AI Security Risks as Autonomous Agents Outpace Controls

Person interacting with computer in modern Singapore office setting.

Nine in ten Singapore organisations experienced at least one AI-related security issue in the past year, the highest rate of the 10 markets in Illumio’s research.

Illumio research: exposure, detection and enforcement gaps in Singapore

Research from Illumio finds Singapore reporting a “sharp and structural rise in cybercrimes” as organisations adopt AI agents. The pattern Illumio describes is threefold: greater exposure, better detection, and weaker controls. The company’s regional senior director, Andrew Kay, told iTNews Asia that “the authority organisations give AI is outpacing their ability to see and constrain what it does. Singapore is simply experiencing that gap earlier than most.”

Illumio’s findings include several concrete markers of that gap: more than half of Singapore organisations enable AI to act autonomously in production; four in 10 assign AI agents privileged identities; and a similar number have discovered an AI agent or tool connected to systems without security approval — the highest rate among the markets Illumio surveyed. The report also found only about 4 in 10 Singapore organisations can see and audit every action an AI agent takes, compared with 49 percent globally.

Agentic AI reaching systems and the risk of legitimate-credential movement

Kay warns that an autonomous agent “with legitimate credentials and live connectivity, can move through an enterprise faster than the organisation's policies can anticipate.” That dynamic changes the attack surface: authorised access becomes a pathway for rapid lateral movement. Illumio’s account goes further, noting that this is not merely a detection artifact — “that is not a detection effect. It is AI reaching systems that no one authorised.”

The broader context Illumio and iTNews describe is rapid adoption across high-connection sectors. The Economic Survey of Singapore, cited in the reporting, shows AI adoption concentrated in information and communications, electronics, professional services, finance and insurance — sectors that often span hybrid cloud, legacy core systems and third-party integrations, each connection presenting “a potential path for an agent,” Kay said.

IMDA’s Model AI Governance Framework for Agentic AI and organisational accountability

Regulatory guidance is clear on accountability: “while agents may act autonomously, human responsibility continues to apply,” according to IMDA’s Model AI Governance Framework for Agentic AI as reported in iTNews Asia. The framework states organisations deploying agents — and the humans overseeing them — remain accountable for agents’ decisions and actions.

Kay condenses that requirement into actionable stewardship: “Every agent in production should have a named owner who can answer three questions immediately: what it is permitted to access, what it has actually accessed, and how to shut it down. Without those answers, accountability exists in name only.” He adds that many policies fail because they are “written around intent rather than reach.”

Technical controls Kay recommends: microsegmentation, shared maps and enforced boundaries

To translate policy into enforceable constraints, Kay argues organisations must move “policy from documentation into the environment itself.” His recommended playbook emphasises containment and enforced boundaries rather than detection alone:

  • Maintain a shared view of the environment so security, IT and AI teams “work from the same map.”
  • Keep security from day one: “No agent should reach production without a defined scope and security sign-off,” noting that four in 10 Singapore organisations have seen agents bypass that step.
  • Measure reach, not alert volume: focus on how far a compromised agent could move and how quickly it can be stopped.

Kay identifies microsegmentation as the practical mechanism to enforce boundaries: restrict communications so a single compromise cannot spread unchecked, and design systems so affected elements can be isolated in seconds rather than days. “What matters is what happens next,” he said — not whether an intrusion occurs, but whether the organisation can prevent a localised incident from becoming an enterprise-wide breach.

What this means for financial services, security teams, and regulators

Financial services already lead on traceability: Sumsub’s APAC AI Governance Benchmark, produced with the Singapore FinTech Association, ranks financial services first on AI governance, with 68 percent of firms maintaining audit trails of AI decisions. But Illumio’s research warns that rapid deployment in IT and software services could outpace governance.

Security and IT teams face the immediate task of mapping dependencies and enforcing reach-based controls; Kay’s prescription is operational rather than rhetorical. Regulators and policy bodies — exemplified by IMDA’s framework — must ensure that accountability requirements are matched by mechanisms to maintain meaningful human control and oversight.

Singapore’s present position, Illumio suggests, is revealing: adoption and authority have moved faster than enforceable constraints. The practical choice the research leaves on the table is clear — organisations can either treat agentic AI as an adviser, constrained by policy enforced in the environment, or as an operator whose reach must be limited by design; the difference, Illumio’s experts warn, determines whether a single compromise remains a local incident or becomes a systemic breach.

Source: iTNews Asia — Why Singapore is most vulnerable to the rising wave of AI security risks