S’pore will take a proactive, practical and collaborative approach to agentic AI: Josephine Teo

Artificial intelligence (AI) systems capable of autonomous actions present unique challenges for society, raising critical questions about accountability when technology malfunctions and when humans lose control over it.

“Who is accountable when agentic AI malfunctions?” asked Minister for Digital Development and Information Josephine Teo on Oct 22.

Whether it is agentic AI or quantum computing, which harnesses the laws of quantum mechanics to solve complex problems, both require something new from policymakers, she said.

Singapore will take a proactive, practical and collaborative approach to govern such emerging technologies – even before their implications can be fully predicted, said Mrs Teo.

She was speaking at the opening address for a panel on AI on day two of the Singapore International Cyber Week conference, where she announced three new initiatives to guide the use of AI and quantum computing.

These initiatives include an update to the Cyber Security Agency’s (CSA) guidelines on securing AI systems, an agreement with major technology companies to share AI-driven intelligence on cyber threats, as well as the launch of quantum readiness and safety documents for public consultation.

AI is already being used to supercharge disinformation and scams, and it has also raised fears about systemic unemployment and cyber warfare.

Quantum computing, meanwhile, with its ability to crunch numbers at incredible speeds, can break current encryption and threaten the foundation of digital systems, potentially altering the way people think about trust.

Both technologies demand immediate action from governments, said Mrs Teo.

She suggested three objectives at this point: Build trust with citizens, ensure that frameworks and tests for AI and quantum computing are relevant and robust, and take timely action.

To build trust with citizens, governments need to establish practical frameworks for testing, validation and accountability for risks associated with AI agents and quantum computing, even when they do not exercise control. And this must be done before the systems are deployed at scale, because it may be too late to address the dangers by then, she said.

For the second objective, she said safe spaces must be provided for experimentation with appropriate guard rails, so that the tools and frameworks developed are applicable to the real world.

On the third, she said policymakers need to act early.

“In several areas, we know the costs of not having acted early enough – the digital divide, misinformation, disinformation, online harms and scams, for example. Let us try not to make the same mistakes with agentic AI and quantum,” she said.

In Singapore, for instance, agentic AI offers tremendous potential to enhance public service delivery, anticipate citizens’ needs and boost national cyber security, but it is crucial to identify risks systematically before granting AI systems autonomy, she said.

She also said assurance must be made practical and measurable.

To this end, the Government will build on the CSA’s Guidelines and Companion Guide on Securing AI Systems to include agentic AI networks. Through real deployments, government agencies also “learn by doing”, said the minister. 

The GovTech-Google Cloud sandbox initiative, for instance, allowed public agencies to test and evaluate the risks of Google’s latest agentic AI capabilities and develop mitigation measures.

“By observing how these systems behave – and sometimes fail – we learn what guard rails are truly needed,” she said.

The Government is taking a sector-specific approach to risk management, ensuring that governance measures are proportionate to the potential dangers, she said, adding that humans remain ultimately responsible.

Meanwhile, it has also moved to help organisations manage the risks of quantum technology.

While there is growing awareness of the threat posed to current cryptographic systems, few organisations have embarked on quantum-safe migration, or the process of moving to systems that can resist powerful quantum computers and protect data.

This is likely because of uncertainty over developments and the lack of specific guidance, a gap the CSA will address, she added.

The agency will launch two resources for public consultation.

They are the Quantum Readiness Index, a tool that helps organisations understand how prepared they are for threats to encryption; and the Quantum-Safe Handbook, which provides guidance, particularly for critical information infrastructure owners and government agencies, to prepare for the transition to quantum-safe cryptography.

Successful governance of these technologies will require international cooperation, said Mrs Teo.

This is because a breakthrough in quantum computing anywhere will affect encryption everywhere, and a vulnerability in one country’s systems can cascade globally.

She suggested that international cooperation must shift from principle to practice, such as through governance frameworks that operate across different systems and countries. In this way, companies can test their agent AI systems once and comply globally.

Singapore is taking proactive steps to address the challenges of agentic AI internationally by releasing for public consultation an addendum to CSA’s Guidelines and Companion Guide on Securing AI Systems.

At the Asean Ministerial Conference on Cybersecurity on the same day, Mrs Teo also announced a collaboration between Microsoft and the Asean-Singapore Cybersecurity Centre of Excellence in Singapore, which will focus on the use of AI for cyber security, among other initiatives.

CSA will also be signing memoranda of cooperation with major technology companies, including Google, Amazon Web Services and TRM Labs, to share AI-driven intelligence on cyber threats and enable joint operations against malicious activities.

She cited Singapore’s partnership with Google on the enhanced fraud protection feature within Google Play Protect, saying it has already demonstrated tangible benefits by blocking 2.78 million malicious app installations across 622,000 devices in Singapore as at September 2025.

During the panel on AI, panellists called for more partnerships as they are key to developing a proportionate way of managing vulnerabilities.

Mr Andrew Elliot, deputy director for cyber security innovation and skills at the British Department for Science, Innovation and Technology, said: “In the struggle to keep up, we’ve also got to make sure that we are driving innovation at the same time and driving adoption as our primary objective.”

Google vice-president of security engineering Royal Hansen said that putting in guard rails will boost adoption, likening standardised guard rails to brakes on a car.

“(They) aren’t there to stop you from doing what you want to do with the car, it’s actually so you can adjust and go the right speed for the right circumstances,” he said.

“Everyone’s going to use (agentic AI) in a slightly different way, but they need those guard rails to feel comfortable that they can learn.”

Society will need to have experience with the technology and be familiar with how it works to make better decisions, he added.

Meanwhile, Singapore’s chief AI officer, Dr He Ruimin, who is deputy government chief digital technology officer at the Ministry of Digital Development and Information, said AI agents can also help to mitigate the risks of the technology.

“These agents do not just stand alone. You can have other agents supervising it... that could be more secure than what an individual agent does,” he said.

Ultimately, all the stakeholders will have to come together to define what is deemed “good enough” for an AI agent, said Ms April Chin, co-chief executive of Resaro, a company that builds AI testing tools to evaluate the performance, safety and security of the technology.

“It always boils down to the trade-off, right? There is no perfect AI out there in the world.”

Source: The Straits Times © SPH Media Limited. Permission required for reproduction.

MAS to announce in November measures to help listcos deliver greater shareholder value: Chee Hong Tat Non-consensual sexual content, promotion of risky actions among severe online harms in Singapore