If I withdraw consent, is my data removed from an AI model that already learned from it?

This is the most honest gap in the framework. Withdrawal stops further processing and should trigger erasure of the underlying records. But a trained model does not store records the way a database does; the information is absorbed into statistical weights from which one person's contribution generally cannot be cleanly extracted without retraining the model. Machine unlearning is an active research field rather than a settled compliance solution. MeitY's India AI Governance Guidelines acknowledge that consent and purpose limitation need greater clarity as applied to AI. Until that arrives, the practical answer is to minimise what enters training in the first place.