Responsible AI Checklist Before Deploying a Model

Before deploying, ask: Is the training data representative and free of harmful proxies? Have you measured performance across demographic subgroups? Is there a human-in-the-loop for high-stakes decisions? Do you have a model card documenting limitations? Is there a rollback plan if the model behaves unexpectedly in production? Responsible AI is an ongoing practice, not a one-time checkbox.