The premise of the experiment is that we have already established that the AI is dangerous. Even if you weren't sure, you should always side with caution and not let it out.