Because the Wald statistic assumes a symmetric distribution that is poorly approximated by the standard normal in the tails.

Because the maximum likelihood estimator for $\beta$ is biased towards infinity.

Because the Fisher information matrix becomes singular.

Because the Wald interval is always too narrow to maintain nominal coverage.