There are a couple of related questions.
First we know (experimentally) that light couples to geometry (gravitational and cosmological red shift, bending of light rays). Then we know (theoretically) how classical electrodynamics can be coupled to gravity; please refer to
http://en.wikipedia.org/wiki/Maxwell's_equations_in_curved_spacetime.
Regarding quantization it becomes rather difficult. There are two steps: i) quantize electrodynamics of classical, curved spacetime; ii) quantize gravity, too.
Regarding i) I am no expert, but there are a couple of well-known results, the most famous one is Hawking radiation which shows how spacetime geometry affects the QFT vacuum. Unfortunately all papers I known do not discuss this effect using photons but massless scalar particles b/c math is much simpler and b/c the expectation is that the main results are not invalidated when going from spin 0 to spin 1. But honestly, I expect severe difficulties in gauge fixing for non-trivial geometries, and even more problems for renormalization.
Regarding ii) this is related to the problem of quantum gravity which us purely understood even w/o taking other fields into account. But as far as I can see nearly all research programs - ignoring string theory and supergravity for a moment - indicate that what we would call a graviton (a quantized plane or distorted wave on a classical background geometry) is not the right entity to be used for quantization of gravity. The basic physical reason is that this approach breaks down when going into the deep quantum gravity regime where this split into classical spacetime + quantum fluctuations becomes meaningless as the latter one must not leave the classical geometry unaffected.
So my conclusion is that a) yes, quantum d.o.f. of the el.-mag. and the gravitational field do interact (and we already see this on the classical level) but that b) using entities like gravitons becomes meaningless and must be replaced by something else.