Singapore flags context-based GenAI risks in global pilot
One of the key findings is that GenAI risks, such as hallucinations, content safety, and data leakage, are highly context-dependent.
Generative AI (GenAI) risks, such as hallucinations, content safety, and data leakage, are highly context-dependent, the Infocomm Media Development Authority (IMDA) and AI Verify Foundation (AIVF) found.
Through its Global AI Assurance Pilot, the IMDA and AIVF recommended that organisations include subject matter experts throughout the AI system development and deployment process. The pilot marks the world’s first coordinated technical testing of GenAI in real-world settings.
Based on insights from the pilot, IMDA also introduced the Testing Starter Kit for GenAI Applications. The kit offers step-by-step testing guidance for businesses and outlines risk categories including hallucination, undesirable content, data disclosure, and vulnerability to adversarial prompts.
In a related move, AI Singapore (AISG) announced a Memorandum of Understanding with the United Nations Development Programme to extend AI literacy to six developing countries.
The partnership will scale AISG’s AI for Good programme, which began in 2024, with the aim of closing the AI skills divide in underserved communities.