Probably ask the maintainers to consent and add some blinding so that the patches look otherwise legitimate.

Ask about this upfront, get consent, wait rand()*365 days and do the same thing they did. Inform people immediately after it got accepted.

Ask Linus to approve it.

This is a good question. You would recruit actual maintainers, [edit: or whoever is your intended subject pool] (who would provide consent, perhaps be compensated for their time). You could then give them a series of patches to approve (some being bug free and others having vulnerabilities).

[edit: specifying the population of a study is pretty important. Getting random students from the University to approve your security patch doesn't make sense. Picking students who successfully completed a computer security course and got a high grade is better than that but again, may not generalize to the real world. One of the most impressive ways I have seen this being done by grad students was a user study by John Ousterhout and others on Paxos vs. Raft. IIRC, they wanted to claim that Raft was more understandable or led to fewer bugs. Their study design was excellent. See here for an example: https://www.youtube.com/watch?v=YbZ3zDzDnrw&ab_channel=Diego... ]