Illustration of two people seen from behind at a desk, looking at a large monitor where an agent inside a glass cube connects through a branching path of steps to calendar, database, and payment card icons, with one step paused at a raised-hand checkpoint. A second screen shows logs, and a sealed folder lies on the desk.
AI-generated editorial illustration of outside testers examining an agent's tool calls and an approval checkpoint in a sandbox. Conceptual scene; it does not depict IMDA, the AI Verify Foundation, any accredited firm, or the Starter Kit's methods. Reuse license pending. Audit Commons (AI-generated editorial illustration) · Original AI-generated editorial illustration; reuse license pending

Three Initiatives Announced in Singapore

Singapore's Infocomm Media Development Authority (IMDA) announced three initiatives in a press release dated October 14, 2026: a Starter Kit for Testing Agentic Systems for Safety and Reliability, an AI Tester Accreditation Programme (AI TAP) under the AI Verify Foundation (AIVF), and the Frontier AI Resilience and Security Taskforce (FAIRS). IMDA made the announcements at the AI Assurance Exchange: The Agentic Edition, an event organized by AIVF and SGTech, and says FAIRS will publish a report with guidelines on AI-driven cyber threats to the infocomm and media sector by the first half of 2027.

IMDA calls the Starter Kit (Version 1.0, October 2026) a practical reference and a living document. The kit sets four steps: identify relevant risks, create realistic tests that limit real-world harm, run tests and analyze logs, and monitor and re-test continuously. It names agent metrics such as tool use accuracy, handoff accuracy, and human-in-the-loop rate, and notes that third-party testers may have only black-box access. The sources we read describe neither the Starter Kit nor the accreditation program as a legal requirement.

What the Accreditation Covers

IMDA says AI TAP applicants undergo technical, operational, and financial evaluations of competency, track record, delivery capability, and financial sustainability. The first five accredited firms are AIDX Tech, Knovel Engineering, PwC Singapore, Resaro, and Vulcan. According to IMDA, accreditation lasts one year; AIVF's program page says 12 months, with renewal required.

AIVF's Accredited Testers page, as we read it on October 18, describes the assessed competence as testing Generative AI systems by benchmarking, red teaming, or both, across five risk areas including hallucination and inaccuracy, data leakage, and bias and fairness. AIDX Tech is listed for benchmarking only, Vulcan for red teaming only. AIVF says it does not certify that an accredited firm's testing is independent or that tested systems are completely safe and reliable. Neither AIVF page mentions agents.

Audit Questions for Hiring a Tester

Our editorial questions: does the accredited scope match the agent under test, given that the AIVF pages describe Generative AI testing while the Starter Kit adds tool use, trajectories, and human approvals? Will the tester receive logs and traces, or only black-box access? AIVF says assessment reflects the agreed scope at evaluation time and leaves risk acceptance to the buyer. Who accepts residual risk, and what triggers re-testing after the agent changes? See also our agent action worksheet and our AICPA brief on auditor responsibility.