it turned out that the ADC card is not broken. Instead it's a common-mode range problem of the inputs.
As we are using several PDs on the table powered by individual (non-grounded) power supplies (Thorlabs photodiodes), these signals are floating around.
And turns out that for some reason since a couple of days the common-mode voltage was sometimes larger than the input can handle (+/-12V) and so strange things happen to the signals from time to time.
In order to fix that i added 1MOhm resistors from the negative input to analog ground of the adc-card to prevent that input from floating around. The negative input is usually ground as all our signal are single-ended referred-to-ground signals; there are no differential transmitted signals for those devices) and so this should work fine. I only added one resistor per device connected to the DAQ, meaning the temp box with four signals has only one (the first one) connected to ground as all other ones have the same ground and so they can't floaf different anymore.

That fixed all problems...