The alternative is to treat everyone as the null hypothesis population, or administer the treatment to everyone and then be unable to determine if it's effective.

Both of these are strictly worse, since neither of them can result in the discovery of an effective treatment.