Semiclassical gravity does violate the equivalence principle.
For example, consider the effect of QED in a gravitational field on photon propagation, discovered by Drummond & Hathrell:
- Drummond, I. T., & Hathrell, S. J. (1980). QED vacuum polarization in a background gravitational field and its effect on the velocity of photons. Physical Review D, 22(2), 343, doi:10.1103/PhysRevD.22.343.
The effect could be understood in qualitative terms as a vacuum polarization from virtual electron–positron pairs $e^-e^+$. These pairs would experience tidal effects from anisotropic curvature tensor with a characteristic length scale, the Compton wavelength of the electron. As a result electromagnetic field would now gain coupling to the curvature tensor. Therefore a photon propagating in a curved spacetime would have a velocity dependent on its direction and polarization and differing from the “normal” speed of light $c$. We now have gravitational birefringence and the possibility of superluminal photons (such FTL photons would not violate causality).
Of course, such effects in a real astrophysical situations would be immeasurably tiny.
For example, in the gravitational field of a Schwarzschild black hole the characteristic relative difference between velocities of different polarizations
$$
\epsilon \approx \frac{\alpha}{30 \pi } \frac{r_s ƛ^2 }{r^3} ,
$$
where $r_s$ is the Schwarzschild radius of a black hole, $ƛ$ is the (reduced) Compton wavelength of the electron, $\alpha$ is the fine structure constant. For a black hole of 1 solar mass this is about $10^{-36}$ near the horizon.