lets examine this one point at a time

I know it's not impossible to get voltage that high with all the coils and capacitors in this thing but I find it
hard to believe it would happen without being specifically designed to do that.
it is impossible to get a dc voltage that high between two points connected by a low impedence unless a massive current is
flowing , circuit theory 101
I've tested it several times and my multimeter was showing almost 2KV last night.
qed the fg[psu frame gnd] is not connected to the 12 gnd
Could there be some capacitance or something that's confusing the multimeter?
no
The main PCB has 4 large capacitors between GND and 12V. I've tried the multimeter in DC and AC mode.
AC mode shows 0V between ground points.
as you would expect, the voltage is basically a static build up
In the PSU the secondary is *almost* completely isolated from the primary.
as you would expect
After the 240V AC is rectified, the negative goes through what looks like a very large blue ceramic disc capacitor
to the negative on the secondary. Could this be causing issues as it's essentially linking half cycles from
live and neutral to my PCB ground.
no , the cap is for rfi suppression and usually 2.5kv rated and a few hundred pf in value
you may also have similar caps on the secondary rails, its also possible to have a MOV between the secondary rail
and fg to ensure psu isolation voltage rating is not exceeded

The problem is also strangely intermittent. It doesn't seem to happen for the first few minutes after being powered
up.
as you would expect the measured voltage is basically a static build up from switching transients and takes time to
accumulate
My PCBs work fine and the multimeter reads 0V between all ground points.
fg is not connected to dc gnd ,it simply cannot be .
it is impossible to get a dc voltage that high between two points connected by a low impedence
without a massive amount of power being consumed .
While in that state, if I put my multimeter between the 0V and 12V outputs on the PSU it reads 12V
and my PCB starts working as normal. As soon as I disconnect it the PCBs start resetting again.
i would guess that the meter impedence probably bleeds off the static buildup in some way
although "the 0V" mentioned is ambiguous