Singapore’s efforts to build international coalitions around safer artificial intelligence (AI) could help establish practical safeguards for the rapidly evolving technology, particularly as the emergence of agentic AI gives systems greater ability to access data, make decisions and execute actions independently.
According to Abhishek Kumar Singh, Head of Security Engineering, Singapore at Check Point Software Technologies, Singapore may not be able to shape global AI rules through its size, but it can influence the direction of AI governance through trust, technical credibility and practical implementation.
He expects emerging coalitions to initially focus on voluntary frameworks, common testing approaches and shared safety practices rather than binding rules.
Over time, however, wider adoption of these safeguards could influence international standards, procurement requirements and regulation.
“Singapore does not need to dictate the global standard to influence it — it can help build the model that others choose to adopt,” Singh said.
Human Control Must Be Designed Into AI Systems
The rise of agentic AI is also challenging conventional approaches to human oversight.
Rather than expecting people to react as quickly as autonomous AI systems, Singh said organisations should establish clear boundaries, authority and accountability in advance, with automated security controls enforcing those boundaries at machine speed.
For high-impact, sensitive or irreversible actions, human authority should remain the final control.
The approach effectively places humans in charge of defining what an AI system is permitted to do, while technology ensures those limits are enforced in real time.
“Humans set the boundaries; technology enforces them at machine speed,” Singh said.
Guardrails Should Not Become A Barrier To Innovation
Singapore’s challenge will be to accelerate AI adoption while ensuring safeguards are proportionate to the risks involved.
Singh likened effective AI guardrails to the brakes on a high-performance car, arguing that they should allow organisations to move faster safely rather than prevent experimentation.
Under such an approach, lower-risk applications could be given greater room for experimentation, while stronger safeguards would apply when AI interacts with critical infrastructure, sensitive information or consequential decision-making.
Such a framework could ultimately strengthen Singapore’s position as a destination for technology companies and investors, rather than undermine it.
Clear and predictable rules, Singh said, can give companies greater confidence to build, test and scale AI systems.
The greater threat to innovation, he argued, is not the existence of guardrails but regulations that are unclear, unpredictable or constantly changing.
Agentic AI Changes The Security Equation
The biggest shift with agentic AI is the move from systems that advise to systems that act.
As AI agents gain the ability to access corporate systems, make decisions and execute actions, the consequences of a security failure could become significantly greater.
This means organisations will need to move beyond simply protecting AI models and applications towards controlling what AI systems are authorised to access, decide and execute.
For Singapore, this is particularly relevant as businesses and critical infrastructure operators explore increasingly autonomous AI applications.
Singh said Singapore has established a strong foundation in AI governance, although operational readiness will vary across organisations.
For critical infrastructure, he advocated a progressive approach involving tightly controlled deployment, continuous testing and clear human authority over consequential actions.
The objective, he said, should not be to pursue maximum autonomy, but rather safe, bounded and accountable autonomy.
As agentic AI becomes increasingly embedded in business operations, the ability to define and enforce these boundaries could become a critical component of enterprise cybersecurity and AI governance.
