OK - I had some time to test tonight.
I checked the voltage between the AttoPilot ground and the actual ground wire that plugs into the power supply - its around 2mV.
Then I checked the voltage between the power supply ground wire and the USB shield wire - about 7mv. This is with the arduino disconnected from the AttoPilot sensor (because it could mask the ground loop).
THEN... I checked the voltage between the power supply ground wire and the arduino ground - and its 32mv. YIKES! And this makes no sense - I thought the USB shield and the ground should be exactly the same, but I guess not!
I took a wire with alligator clip ends, clipped one to the main ground wire, and another to the arduino ground pin, it drops the voltage difference down to 2mV. When I do this the accuracy of the reading gets better, though. Actually power supply usage is around 1.2A (I removed some stuff from the PC to make sure they weren't causing issues) and when I connect the ground wire the sensor reading goes from 0.4-0.5A up to .7-.75A.
Actual sensed voltage on the arduino is now 0.03v, my multimeter on it says 0.033v. The expected reading should be more like 0.045mv, so it definitely now seems like the ground issue is resolved, but I think my AttoPilot sensor is bad.