That’s fine as long as I can identify and reject any Mythos derived patch as being irreproducible.
Why would you refuse to use a patch that deals with a valid PoC exploit?
If a random contributor posted an explanation of an exploit, showed it worked in an executable way, presented a patch and you could see that the exploit no longer worked - would you refuse to use the fix until the contributor showed how they figured it out?
When an ungated (or even abliterated) public model can repeatedly, easily, and accurately embarrass Anthropic’s models, that might change.