An improper prior is non-negative but does not sum to $1$ and so is not a proper probability distribution. However, if you can verify that the improper prior yields a proper [[posterior]] distribution, the inference can be considered valid.