My idea is to use a sort of layered CAPTCHA model (4 layers - from which layer 0 is no captcha) with difficulty as primary focus.
My first focus beside stopping bots is to make the visitor more and more aware of what he's doing.
It will give the visitor more think work than "type-in" or "watch-and-type-in" work and this verifies the visitors knows what he's doing(eg. limitating unnecessary requests).
After many years of working in the information security industry, I have yet to come across such an implementation. This doesn't mean it can't be suitable for certain use cases, but it definitely isn't an established design pattern.
I have seen a decrease in the usage of CAPTCHAs during the last years, because UX experts hate them. CAPTCHAs are mostly used as last resort (e.g. after 10 failed login attempts) or implementations that aim at less UX impact (e.g. reCAPTCHA V2) are used.
But there is another mechanism known in the security world, that might fit your need. It is called step-up authentication. This pattern basically forces a user to re-authenticate with the same or a stronger form of authentication, before he can perform a high-risk task. This achieves what you are looking for: It verifies that the visitor knows what he's doing.
A good examples for this approach is deleting a repository at GitHub. Before you can do this, you must re-enter your password and type out the name of the repository, which you selected for deletion. That way the user is forced to stop and think again, before making a crucial mistake by deleting the wrong repository.
Another example are critical operations which require MFA before they allow execution. You may be able to log into your online banking account with username and password to check your balance, but if you want to perform a transaction, you must enter an additional PIN sent via SMS as a second factor.
You have to evaluate, if a ramp-up authentication would solve your problem in a more elegant way are you really do require CAPTCHAs. Please take into account, that CAPTCHAs are usually not considered user friendly and may have an impact on your customer satisfaction.
This is more of a User Experience question. But if you want to make sure the user is aware of what they are doing before performing a destructive operation and decide to do that by making that operation intentionally difficult to perform, then you should design those hurdles in a way which makes them more aware of what they are about to do. A regular captcha would rather be a distraction, because it usually has no relation to what the user is about to do.
Example: Our company has an inhouse solution for rolling out remote software configuration changes. One administrator intended to uninstall an application on one PC, but misclicked and accidently uninstalled it on a couple thousand PCs. It was a very important application, so it caused a lot of disruption. Our plan to prevent that from happening in the future includes that the remote configuration program now tells the user how many PCs will be affected by their change, and if it's more than a hundred it asks them to type that exact number into an input field before the command gets executed.
