Initiative overview
The Global AI Assurance Sandbox was established following the Global AI Assurance Pilot, which IMDA and the AI Verify Foundation launched in February 2025. By May 2025, the pilot had paired 17 AI deployers with 16 specialist technical testers from around the world. Participants gained practical insight into both what to test and how to test GenAI applications in real-life operational contexts. Following this success, the initiative was formalised as an ongoing Sandbox in July 2025.
The Sandbox serves two primary groups: builders or deployers of GenAI applications seeking guidance and access to specialist testers, and specialist technical testing vendors looking to validate their methodologies, connect with potential customers, and contribute to emerging standards. Applications eligible for testing must involve a large language or multi-modal model, be live or intended for live deployment, and make a net new contribution to the AI Verify Foundation and IMDA's body of knowledge. Testing engagements last up to three months and result in a published case study or report.
Risk dimensions considered during testing, drawn from IMDA's Starter Kit, include hallucination, undesirable content, data disclosure, vulnerability to adversarial prompts, impacts on safety and health, lack of human oversight, and breaches of regulatory or compliance requirements. The Sandbox does not provide a software testing environment or regulatory approval, and explicitly focuses on the application layer rather than underlying foundation models.
