Trai says it is important to regulate specific AI use cases that may have a direct impact on humans within a risk-based framework