If the significance level $\alpha$ is decreased (e.g., from $0.05$ to $0.01$), what happens to the probability of a Type I error?