This is as far as my lay understanding goes: the idea was that you could better handle potential outbreaks by creating potential outbreaks in a lab, and studying them there. If you have bad priors ("we have perfect protocols, because we've never caused a global pandemic") I could see how you'd get to this being a net positive: there will be another pandemic, probably involving one of these, so let's study it before it's a pandemic.

I've thought of the intent as a little like using a fuzzer in security research, although that might not by a great analogy.