Electronic – Smart power switch works fine on a bench but fails to disengage the intended load on location

I'm using the Infineon BTS117 smart low side switch to drive a DC-DC power regulator in my car. Here's the schematic of the switch board:

^{simulate this circuit – Schematic created using CircuitLab}

I don't think all the stuff around BTS117 is relevant to the issue, but then, I'm not an electrical engineer so I thought I'd better lay everything out.

I was testing the device in the car a few days ago and it was working properly, but I couldn't finish the work because one of the input leads was missing. So I took the device home and soldered the lead (just a small piece of wire). Next morning I put it back in the car, but it was no longer working. It wouldn't disengage the load with both inputs floating. As soon as I connect the positive terminal, the load (which, I remind you, is a power supply) would power up. The said PSU is able to work with as low as 4V input according to its specs.
I can't recall for sure (it was 2 days ago), but I believe I did try grounding the gate of BTS117 to the ground, and it made no effect.

I took the switch board home and tested it with a 12V 5W light bulb for a load. It again worked perfectly. Connected a milliammeter in series with the load and measured 0 mA with inputs floating at the same 12V across the load (and hence the switch).
Also tried measuring the channel resistance with inputs floating and power supply disconnected. My DMM has the upper limit of 2 MOhm and that wasn't enough for it to sense any conductance.

Any idea what can possibly be going wrong when I put the switch in the car? Why does it conduct when it should be off?

Update: I have once more experienced the faulty behavior, and this time the switch got considerably warm, i. e. it didn't turn on all the way and had decent voltage drop across it. I have disassembled everything and reassembled back again half an hour later; the problem went away and has not repeated since.