You may remember this work from last year:
What it showed was the following: if you take someone who began playing Pokemon from an early age (5 to 8 years old), who continued on, playing some into adulthood, and you show them Pokemon pics, a certain region of their brain will light up; and the location of this region is more-or-less consistent across subjects. The region doesn't light up for other cartoon or comic characters -- it is specific to Pokemon. Furthermore, novice players, and people who hadn't played at all, don't show the brain response.
The response is probably reliable enough to where you could pick out whether someone was a Pokenmon player with reasonable accuracy. I wouldn't be surprised if a good Machine Learning algorithm could do it with 95%+ accuracy.
Now consider the case of a spy, who goes to work in an intelligence agency of some kind, and sees the exact same buildings, murals, decorations, and so on, every day (or... 5 days a week), for many years. They might not have begun the exposure to the stimuli while young; but the adult brain is somewhat plastic, so we can expect regions of their brain to become devoted to it. And... the location is probably going to be roughly consistent across all the spooks working there.
Now suppose one of these spies is in a foreign country, and is captured. A wearable BCI is placed on their head, and they are shown trigger images of where they worked... and their brain automatically lights up in response. They can't shut it off. If they see the stimuli, their brain will respond.
More generally, any kind of "secret" one wishes to keep, that involves a stimuli one is exposed to over and over again, might be extractable, even decades later. And, probably, if one is willing to settle for slightly less accuracy (say, 90% accurate, instead of 95%+), one can even extract secrets about things one is only exposed to for a short time.
Just imagine the implications of that!