The Pulse
Singapore Launches Agent Test Kit, Accredits Five AI Testers
Singapore’s IMDA launched a starter kit for testing AI agents and their actions on October 14, 2026. AI Verify Foundation named five firms accredited to test generative AI systems within defined scopes.
AI.info Team · Written with AI ·
On October 14, Infocomm Media Development Authority chief executive Ng Cher Pong announced a new testing guide for AI agents, alongside the accreditation of five AI testing firms. The starter kit asks organisations to inspect not just whether an agent reaches the right result, but what it does on the way. Singapore also announced a task force focused on cyber threats tied to frontier AI.
Test the steps, not just the answer
AI agents can use tools, reach external services and sensitive information, and make changes while carrying out a user’s request. A correct final answer, on its own, does not show whether the agent stayed within its permissions or took an unauthorised action along the way. “When we give AI more agency, we need greater assurance over how that agency is exercised,” Ng said at the AI Assurance Exchange: Agentic Edition.
The first version of IMDA’s Starter Kit for Testing Agentic Systems for Safety and Reliability lays out four stages: identify risks, design tests, analyse results and activity logs, then monitor and retest the system. The guide directs testers to examine the agent’s tools, the information it accessed, its decisions, whether it respected assigned boundaries, and whether safeguards worked as intended. Its focus on an agent’s full sequence of actions distinguishes the task from checking a conventional application’s output alone.
IMDA developed the kit after consulting more than 20 industry deployers and testers and drawing on work through the Global AI Assurance Sandbox and Singapore AI Safety Institute, CNA reported. Ng described the kit as a living document that IMDA plans to refine as testing methods develop. The agency’s January Model AI Governance Framework for Agentic AI had set out broader guidance, including human accountability and controls on agents’ access to tools and data.
Five firms receive scoped accreditation
AI Verify Foundation named AIDX Tech, Knovel Engineering, PwC Singapore, Resaro and Vulcan as the first firms accredited under its AI Tester Accreditation Programme. Their accreditation covers demonstrated competencies in specified testing methods and risk areas for generative AI systems, including benchmarking or red teaming and risks such as data leakage, bias and inaccurate outputs. The scope differs by firm.
The accreditation is not a blanket endorsement of each firm’s services or a certificate that the systems it tests are safe. The foundation says it does not certify that testing by accredited firms is independent, or that tested systems are completely safe and reliable. Ng also said accreditation recognises particular testing capabilities rather than suggesting that any one firm can test every system.
A separate task force targets sector-wide threats
IMDA also formed the Frontier AI Resilience and Security Taskforce, bringing together more than 20 leaders and experts from telecommunications, media, cybersecurity and AI. Its work will cover faster responses to AI-accelerated threats, stronger protection for critical systems as AI becomes more capable of finding vulnerabilities, and better sharing of intelligence about emerging attack techniques. The group is expected to publish practical guidance for the infocomm and media sector in the first half of 2027.
The three measures address different stages of assurance: tests of individual agents, a pool of accredited testing providers, and coordinated sector-level cyber preparedness. Their limits matter as much as their reach. The kit is guidance, while the accreditation applies to documented testing competencies; neither replaces an organisation’s responsibility to set permissions, supervise consequential actions and manage remaining risks.