Explainability must be the baseline for AI in compliance