It feels like you’re making the argument that both random users wouldn’t approve anything in the first paragraph and they would readily approve bots in the fourth.
The reality is most users would probably be fairly permissive but might be delayed in their authorizations (ex they’re offline). If a bot acts enough like a person it probably won’t get caught right away but its likely whoever did let it in will be barred from authorizing people. I’m not saying this is a perfect solution but I would argue its an improvement over existing systems as over time users that are better at sussing out bots will likely be the largest group able to authorize people.
I’d imagine there would need to be an option for whoever was an authorization was made to (the authorizor) to start a DM chain with the requesting account.