Naive cross-validation assumes that examples are i.i.d., which is clearly not the case for time series. Cross-validation is still possible for time series, but not while using all of the data in every split. Instead, cross-validation is performed using slices, as shown in the figure below (from the SKL documentation, here):
-learning-(data-science)/60-Training-and-optimization/Cross-validation/Time-series-cross-validation.png)