For low impedance circuits, voltage noise is more important than current noise.
For high impedance, the reverse is true.
With 2.8 nV/rt-Hz, this is definitely an op amp that is designed for low impedance circuits,
so I would have never chosen to use it in a high impedance circuit.
Thanks for the suggestion. It's clear to me that AD8675 isn't suited for input voltage outside the supply rails. Unfortunately AD8675 shows already phase reversal with input voltage within the supply rails, causing the described latching behavior of a +1 buffer amplifier driven by a higher source impedance. It happens e.g. during power-up without any input source.
Choosing a different amplifier is an option, also limiting the output voltage feed back to the inverting input by a voltage divider.
For input voltages that goes beyond the supply rails, Over The Top amplifiers are more suitable as these amplifiers will not experience this. Do check out LT6015, LT1672, and other OTT products on http://www.analog.com/media/en/technical-documentation/product-selector-card/2PB_6015f.pdf and on http://www.analog.com/en/index.html
I experience phase reversal respectively input latch-up problems in a AD8675 buffer application and found this QA as only reference. Particularly the topic isn't discussed in the datasheet.
Saying no phase reversal as long as the input voltage stays in the specified range is a tautological conclusion. Keeping IVR assures normal operation, thus no phase reversal.
The problem arises if you drive AD8675 in unity gain buffer configuration from a high impedance source. A short spike exceeding the input voltage range seems to be able to latch the amplifier output at the negative rail. Input clamp diodes are pulling In+ down below the IVR, the circuit doesn't recover without removing the supply, or shorting the In+ by a lower source impedance.