At least in places where rule of law exists the use of rubber-hose cryptanalysis is significantly restricted especially for powerful authorities.
Strapping someone into a fMRI with a loop that says "YOUR PASSWORDS FIRST CHARACTER A B C ..." isn't (yet) so obviously restricted.
Rubber-hose cryptanalysis can also, at its extremes, be resisted by denial (potentially at the cost of disfigurement or death) at least by some people. Mind-reading cryptanalysis may require that you commit suicide in advance of being questioned to protect information. They aren't equivalent.
I can picture a Sci-Fi story, where someone claims to have a brain scanning technology that can pull information out of someone's head in this fashion. However, the twist is that the device can't read memories. It can only overwrite memories. So the way the extraction tech actually works, is that a virtual reality memory of torture is projected into the subject's brain, breaking the subject and getting them to divulge the information. Then, after the torture, the torture memories are erased with a scene of their mind being "read."
Of course, this doesn't work 100% of the time, so the startup which produces this new tech tries to use it to cover its tracks.
That is sort of like the old "Scotty with the phaser" explanation for transporters.
FMRI extraction of text has already been performed in a controlled setting. No one has demonstrated it with an adversarial subject in the open literature, yet.
An interesting consideration for the extraction of cryptographic keys is that the extraction could be very noisy and yet still be a massive speedup for a search. If the attacker can try a billion combinations per second they may not need to get that many bits of search space reduction from the subject.
