High accuracy can be deceptive if the model has accidentally accessed target information during training, creating a false sense of predictive power.

Try this: Perform a rigorous feature importance analysis and check for high correlation between predictors and the target to identify potential leakage.