The Wald interval assumes the sampling distribution of $\hat{\beta}$ is perfectly uniform.

The Wald interval relies on the asymptotic normality of the MLE, which may not hold in small samples with sparse data.

The Wald interval is always wider than the Likelihood Ratio-based interval, leading to lower power.

The Wald interval requires the parameter to be a variance, not a regression coefficient.