It's happening with two copies of the same circuit using AD8221 and another instance of the same circuit with AD8421. The power supply has <10mV pk to pk of noise as measured with an oscilloscope, and the bypass caps are close to the pins.
The DC voltage is not predictable - it's often not there at the start of my measurement but when it runs for a few hours, which is necessary for the measurement, the offset can become large enough that my AC signal clips at the power rails.
It's on a custom PCB, 4 layers, with a ground plane below the signals in use in this part of the circuit. I also noticed this problem before designing the PCBs when I was using the chips in breakout boards on a breadboard.
Sometimes when the chips show a large DC offset as soon as I turn the power on I've replaced them and the problem has gone away for a little while, but it always seems to come back. The compliance on the AC signal is 10V, and it's only driving ~10mA max, so I don't know what could be burning the chips.
The frequencies of the AC signal are 300Hz to 30kHz. The AC signals seem to be superimposed on the DC offsets, even at the outputs, unless they would exceed the power rails, so it's not a case of the chips being completely destroyed.
I have posted about this before but more recent comments are not getting quick responses. https://ez.analog.com/message/289856?commentID=289856&et=watches.email.thread#comment-289856