IMDA introduces new AI testing and cybersecurity initiatives

28 October 2026

In a move to strengthen AI safety and resilience, the Info-communications Media Development Authority of Singapore (“IMDA”) announced three new initiatives on 14 October 2026 at the AI Assurance Exchange: The Agentic Edition, an event organised by the AI Verify Foundation (“AIVF”) and SGTech.

As AI systems become increasingly adept and autonomous, they pose risks arising from both malicious actors and system anomalies. The IMDA initiatives address these risks though complementary safeguards spanning governance, safety testing, and cybersecurity. They comprise:

  • the introduction of the Starter Kit for Testing Agentic Systems for Safety and Reliability (“Agentic Testing Starter Kit”);
  • the launch of AIVF’s AI Tester Accreditation Programme (“AI TAP”), with an initial cohort of five accredited testing firms; and
  • the formation of the Frontier AI Resilience and Security Taskforce (“FAIRS”).

Agentic Testing Starter Kit

Agentic AI systems can reason and carry out tasks autonomously on users’ behalf, including accessing sensitive data and modifying their operating environment. To address the risks associated with these capabilities, the Agentic Testing Starter Kit provides practical guidance for organisations beginning to test such systems. It is built upon consultations with government agencies and more than 20 industry deployers and testers, alongside testing experience from the Global AI Assurance Sandbox and Singapore AI Safety Institute.

The Agentic Testing Starter Kit sets out a four-step approach:

  • identify risks, such as incorrect actions and behaviours beyond the system’s permitted scope;
  • design realistic and robust tests that cover a range of agent behaviours while balancing against potential harm;
  • conduct tests and analyse logs to assess performance; and
  • monitor and re-test continuously as systems evolve.

The guidance addresses testing across an agent’s reasoning and actions, using agent-specific metrics such as the accuracy of tool use and handoff, as well as human-in-the-loop rates. It also covers the progressive automation of log analysis for key failure signals, and adaptive testing when access to a system is limited.

This Agentic Testing Starter Kit builds on the Model AI Governance Framework for Agentic AI, which guides organisations to implement agentic systems responsibly. It also expands on the earlier Starter Kit for Testing of LLM Apps for Safety and Reliability, which addresses common risks such as hallucination, data leakage, and undesirable content.

The Agentic Testing Starter Kit will be updated with relevant feedback and new best practices.

AI TAP

The second initiative is the AI TAP launched by the AIVF to develop a pool of competent third-party AI testing providers across sectors, including healthcare, finance, recruitment, and public services. The programme aims to help businesses select accredited providers to assess the safety and reliability of their AI applications. Five firms have so far been accredited under the programme.

Further details on the programme are available on AIVF’s website.

FAIRS

IMDA has also established FAIRS to strengthen the info-communications and media (“ICM”) sector’s defences against AI-enabled cyber threats. Bringing together more than 20 leaders and experts from telecommunications, media, cybersecurity, and AI, the taskforce will examine how the sector can harness AI while managing the associated cybersecurity risks.

Since the ICM sector underpins Singapore’s digital economy with interconnected services, any disruption threatens national security and economic resilience. Robust cybersecurity governance is therefore vital to safeguard these operations and boost Singapore’s competitiveness.

FAIRS is expected to publish a report setting out best practices and practical guidelines by the first half of 2027, focusing on three priorities:

  • Enhancing responses to AI-accelerated threats: Improving network visibility and strengthening response readiness to adapt to the shifting nature of threats.
  • Building resilience against evolving risks: Examining how ICM networks and systems can be reinforced and divided into segments to protect critical services and prevent widespread disruption.
  • Anticipating new threats: Gathering sector-wide intelligence and insights to detect new attack techniques ahead of time and translate findings into practical guidance.

Collectively, these initiatives reinforce the Government’s commitment to ensuring that AI remains trusted, resilient, and reliable, so that individuals and businesses can use AI responsibly and confidently.

Reference materials

The following materials are available on the IMDA website www.imda.gov.sg: