Generative AI Evaluation Sandbox — Singapore's IMDA/AI Verify Testbed for Trusted LLM Deployment

On 31 October 2023, Singapore's Infocomm Media Development Authority (IMDA) and the independent AI Verify Foundation launched a Generative AI Evaluation Sandbox, billed as a first-of-its-kind controlled testing environment for large language model (LLM) applications before wider deployment.

Sixteen organisations confirmed participation at launch: model developers Google, Microsoft and Amazon Web Services; app developers Singtel and OCBC testing specific use cases; external testers Deloitte and EY; and Singapore's Personal Data Protection Commission (PDPC) as observing regulator. Each sandbox use case pairs an upstream model developer, a downstream application deployer and a third-party tester, so evaluation happens across the AI supply chain rather than by a single actor self-certifying. The sandbox is built around a shared Evaluation Catalogue setting out baseline testing methods, aimed at surfacing evaluation gaps in domain-specific contexts and languages and generating input for technical testing standards.

According to IAPP's later reporting, the testing framework spans 11 principles (including transparency, safety, repeatability, fairness and human oversight) aligned to the NIST AI Risk Management Framework, the Hiroshima Process International Code of Conduct and ISO/IEC 42001, and by mid-2025 Standard Chartered had piloted a generative-AI application through the sandbox with independent third-party evaluation. IMDA and the AI Verify Foundation have publicised the participant list, framework and objectives, but no independently published data yet reports pass/fail rates, vulnerabilities found, or measurable changes to deployed systems as a result of sandbox testing.