An ordinary photon travels perpendicularly to the direction of its oscillating E & B vector fields (i.e. $\vec{v} \propto \vec{E} \times \vec{B}$). Let's say $\vec{E}$ is oscillating "in-out" of the page, $\vec{B}$ is oscillating "up down", and so $\vec{v}$ propagates to the right. Now turn on a strong uniform downward gravitational field. The photon bends downward, and now $\vec{v}$ is at some angle with respect to the horizontal and has a downward component. I can think of two possibilities: (1) $\vec{E}$ and $\vec{B}$ are still "in-out" and "up-down", and so $\vec{v}$ is no longer $\propto \vec{E} \times \vec{B}$. Or, (2) we still have $\vec{v} \propto \vec{E} \times \vec{B}$ and so $\vec{E}$ and $\vec{B}$ have rotated as the photon bent and no longer oscillate "in-out" and "up-down."
Scenario (2) seems consistent with Maxwell's equations. But scenario (1) seems consistent with the equivalence principle (i.e. can exactly replace uniform gravitational field with an accelerating reference frame). If I imagine there is no gravity and I watch a right-moving light beam from an upward-accelerating reference frame, I would still see the light beam bend down, but (I think?) I would still see the fields oscillating "in-out" and "up-down," no longer orthogonal to the perceived direction of propagation of the light beam.
Which scenario is correct, and how is it internally consistent? Thanks!