This is related to an issue called root-phishing or superuser-phishing. You can do this with the Windows admin password prompt, the MacOS prompt, or with Linux sudo, as long as you can run code from a user account or edit a single file. alias sudo='sudo ./somethingbad; sudo' I'm surprised you don't hear about this that often. There is no perfect solution, since any visual feedback the operating system can do to make…
So the dialog pops up and says "please press button then enter your password". User needs to press the button for the textbox to appear. This hardware button auto-closes whatever the active application if the active application is not the system dialog.
This way you have a hardware control verifying the dialog isn't a phishing attack.
Of course, attackers could mimic the dialog after the button had been pressed if it were a genuine dialog. But as long as you can train your users into only entering their password after they've pressed the button then hopefully most might realise something was amiss if requested for a password without the button press.