Testing IndiaAI safety commitments in practice
AI Safety Becomes Real Only When It Can Be Tested
The Ministry of Electronics and Information Technology’s latest update on the IndiaAI Mission lists 13 responsible-AI projects, 58 AI Centres of Excellence, and 27 India Data & AI Labs. It also mentions support for 20 sovereign-model proposals: 12 large language models and 8 small language models. The Safe & Trusted AI pillar is therefore moving beyond a slogan and into programme management.
The engineering question Numbers are useful, but they are not evidence of safety. For each project, I would want public test protocols, versioned evaluation results, incident reporting, data-governance records, and a clear owner for corrective action. A model that passes a demonstration but cannot explain its failure modes is still an operational risk. Sovereign capability matters. So does the unglamorous work of measuring what breaks.