Singapore AI Governance Framework GEN-3: Apply baseline safety practices, evaluate models and publish food-label disclosures

GenAI framework, Trusted Development and Deployment: developers and deployers should apply safety practices across the lifecycle (fine-tuning such as RLHF, use-case risk assessment, input and output filters, retrieval-augmented generation to reduce hallucination), evaluate through benchmarking and red teaming working towards standardised safety evaluations, and disclose in a standard way the data used, training infrastructure and environmental impact, evaluation results, mitigations, risks and limitations, intended use and user data protection, with more transparency to government for high-risk advanced models.