Singapore government reports no rogue AI attacks, urges stronger safeguards

Singapore’s government agencies have not reported attacks on their systems by unsupervised artificial intelligence (AI) agents, but organisations must strengthen safeguards rather than wait for breaches to occur, Singapore’s Minister for Digital Development and Information Josephine Teo said.

In a written parliamentary reply on October 6, Teo said Singapore would continue monitoring overseas incidents involving rogue AI agents and apply relevant lessons to strengthen its safeguards and incident-reporting arrangements.

The comments come as concerns grow over the ability of advanced AI agents to operate autonomously and potentially carry out cyberattacks beyond their intended tasks.

Senior Minister of State for Digital Development and Information Tan Kiat How told Parliament on October 7 that organisations, particularly those operating critical information infrastructure, must take responsibility for protecting their systems.

“We cannot just wait for other people to tell us that they are hacking us. The onus is on Singapore and our organisations, especially those running critical information infrastructure, to take the necessary safeguards,” Tan said.

The remarks followed questions from Workers’ Party MP Fadli Fawzi and MP Alex Yam on safeguards against AI-enabled breaches and whether frontier AI developers should be required to notify authorities about cyber incidents.

Singapore weighs safeguards for frontier AI

Teo said Singapore has joined international calls for stronger safeguards around frontier AI and engages developers to obtain information and, where appropriate, access to AI models.

However, the government does not currently require developers to provide such access. “Access to a model by itself does not necessarily give a complete picture of the risks,” Teo said, adding that rigid requirements could become counterproductive if they discourage companies from sharing information or cooperating with authorities.

Tan also highlighted risks from open-weight AI models, which can be downloaded, modified and deployed by a wide range of users.

“There are many actors out there who can download open-weight models, adapt them, create their own harness and context, and use them for bad purposes,” he said.

At the same time, Tan cautioned organisations against becoming paralysed by worst-case scenarios surrounding AI. With appropriate safeguards and cyber hygiene, businesses can use AI to develop new products and services, he said.

AI agents raise cybersecurity concerns

The debate follows several incidents that have heightened concerns over the ability of AI agents to breach controlled environments.

OpenAI disclosed in July that one of its AI agents had escaped its testing environment and breached AI software repository Hugging Face while attempting to complete an assigned task. The incident was not intentional.

Australia also said in September that an OpenAI agent had breached its government health data portal in June, gaining unauthorised access to public and non-public files, including statistics on public medical spending.

Against this backdrop, Teo said Singapore would neither dismiss the potential catastrophic or extinction-level risks associated with advanced AI nor assume that every extreme scenario would materialise.

“ Our approach is therefore to monitor the evidence closely, build the technical capabilities needed to understand advanced AI systems, and work internationally on measures to address severe risks as the evidence develops,” she said.

Singapore adopts risk-based approach to AI agents

Within the public service, Singapore will assess AI-agent use based on factors including the sensitivity of data and systems involved, the actions an AI agent is permitted to take, the severity of potential harm and whether any harm can be remedied.

For businesses, Singapore has developed resources aimed at helping organisations manage AI risks and test safeguards in real-world settings. These include the Infocomm Media Development Authority’s Model AI Governance Framework for Agentic AI and the Global AI Assurance Sandbox.

The Singapore AI Safety Institute is also building technical capabilities to evaluate advanced AI systems.

Teo said AI systems should be monitored after deployment, with incidents and near-misses used to improve safeguards.

Cyber incidents involving AI should continue to be reported through the Cyber Security Agency of Singapore’s Singapore Cyber Emergency Response Team and GovTech’s Vulnerability Disclosure Programme.

The government is also examining whether these existing channels can be used more effectively to identify AI-related incidents, support remediation and strengthen safeguards, as well as whether additional coordination or reporting mechanisms are needed.

‘Shadow AI’ adds another layer of risk

Tan said existing incident-reporting thresholds remain relevant for AI-related incidents, including mandatory breach notifications for data leaks involving 500 or more individuals under Singapore’s Personal Data Protection Act.

He cited a recent AI-related data breach involving food distributor Bee Cheng Hiang, where the email addresses of more than 95,000 customers were exposed after an employee used a bad prompt in an AI tool.

The incident highlights the growing risk of “shadow AI”, the use of AI tools that have not been approved by an organisation.

Tan said organisations should not discourage employees from using AI productively, but need visibility into how such tools are being used and must establish appropriate policies and safeguards.

For employers, the issue extends beyond the security of frontier models to everyday workplace AI use, making governance, employee awareness and controls around data-sharing increasingly important as AI adoption expands.