If Akaike Information Criterion (AIC), Schwarz Criterion (SC) and Hannan Quinn (HQ) Criterion unanimously choose lag 1, that is a powerful choice. It should be implemented; any additional lag is an over-parametrization. Parametric parsimony considerations make that unnecessary.

Generally, I would not worry that the model has too few lags. Parsimony is an advantage, i.e. having less parameters you obtain more precise estimates. Moreover, in a macro-forecasting you usually get better results when taking less lags (even if the model is misspecified).

However, you may need more lags due to specific reasons. VAR(1) models can not generate hump-shaped impulse response. More dynamics may be also needed if you wish to simulate cyclical pattern, etc.