EDIT: What I forgot about with the lithium option is that it also reacts with Oxygen and Nitrogen, but it was a fun thought while it lasted. Thanks to Dmitry for the comment.
If you have access to a manufacturer that can do some tricks with metals, you could do a trial run of a small resistor-type component with an exposed lithium link.
Lithium reacts with water, but not quick or violently enough to melt or explode. I have no idea how "legal" that would be though, because it will outgas. And well, lithium is scary now.
And you will have to "dessicate" the box before revealing the lithium, or it might start oxidising with the existing moisture level in the air present.
Otherwise you could make a leakage sensor strip that triggers a destructive process with a MOSFET or some such.
simulate this circuit – Schematic created using CircuitLab
If the leakage between the traces (make them without Soldermask and possibly ENIG?) starts beating the 22M resistor it will start shorting out the power, draining a lot through the fuse, it will "vaporise" soon enough.
It may be needed to make it a two-stage system, because this can take a while to start conducting through the MOSFET and you don't want it to break, because that is your safety, like this:
simulate this circuit
In this case the MOSFET will only need to start conducting a few mA for the PNP transistor to boost that up nice and strong, making the reaction much quicker.
You can also put a P-MOST in place of the PNP, this will really boost up the reaction, but it might also become "too hair-trigger". If you make the pinch-off resistor (R3) even larger with the P-MOST it might start triggering with tiny little spikes of leakage that you might still want to ignore.
But you have to realise that this will always only "sense" where you put it on the board. The advantage of this is, with a fuse on the input you can riddle the board with shorting-mosfets that will blow the fuse. If the current consumption of your device is low, you can keep the fuse small and you could use SOT323 type tiny transistors, wasting little space.
You will need to tune the component values (and types: PNP/P-MOST) during some tests, you can estimate the initial values for the 22M resistor when you know the distance between the interlocking traces and what kind of water creeps in. I'd advise you to make the interlocking traces the once with the least clearance on the board, that gives you some safety: They'll be the first to start leaking noticeably.