Thanks @attila for the answer. After following your instructions, it turned out that this odd behavior only happen when the AD2's GND is connected to my testbed's GND (my microcontrollers' GND, I'm not sure if this is the right GND to connect to though!). Once, I disconnected the AD2's GND, CH2 works but I lose CAN's partial information from CH1. Example here:
Any idea why would connecting GND results in losing CH2?
Also, is it ok to leave the rest of GND probes and CH1&2 negatives open?