The issue here is that the OP did not abuse anything, tried to correct the mistake once discovered and was 'still' the one being punished for someone else's shady practice.
GitHub is taking the “ban them all and let God sort ‘em out” approach to figuring out if OP is telling the truth.
Otherwise it would be quite simple to write a malbot and then claim innocence because it was the bot doing it, not me.
I think the approach to automation is best when the authority and responsibility always ties back to an individual or group.
I automatically decline the moment I see any app trying to authorize with that scope.
Nonetheless, perhaps this is pointing to Microsoft’s Window’s UAC moment for GitHub.
Bright yellow or red UX with warnings that if you click “agree” then you might as well have given away your computer to a malicious actor.