The OLS estimators are biased.

The standard errors of the regression coefficients are inflated.

The model $R^2$ is always low.

The $F$-statistic for the overall model significance becomes invalid.