What the OP most probably has in mind is the Bohmian version of the EPR paradoxon, which is the most simple version of it using a "two-state observable". As an example take a neutral pion at rest decaying to two photons. Since the pion is a pseudoscalar particle the total angular momentum of the two photons is 0, and they fly back to back in, say, direction ##\pm \vec{k}=\pm k \vec{e}_z##. This prepares an entangled two-photon state of the type
$$|\Psi \rangle=\frac{1}{\sqrt{2}} (\hat{a}^{\dagger}(k,x) \hat{a}^{\dagger}(-k,y) - \hat{a}^{\dagger}(k,y) \hat{a}^{\dagger}(k,x)).$$
You have to consider the ##\hat{a}^{\dagger}(k,j)## as the creation operators of photonic wave packets with polarization ##j \in \{x,y \}##.
Now in the OP's scenario you put detectors at ##z_{1,2}=\pm 1 \text{Ly}## and measure the polarization (say you put a polarizing beam splitter like a birefringent crystal at the places).
It's clear that the polarization states of each photon is completely indetermined, i.e., they are exactly unpolarized, but if the observer at ##z_1=1 \text{Ly}## finds his photon to be ##x##-polarized he instantaneously knows that the observer at ##z_2=-1 \text{Ly}## must find her photon to be ##y##-polarized and vice versa.
Now if you take the "collapse hypothesis" a la some flavors of Copenhagen as a physical process indeed, the polarization of the photon instantaneously changes by the measurement of the other photon's polarization. This seems to be violating Einstein causality since it should take 2 Ly for the signal of the measurement at the one place can influence anything at the other place.
The resolution of the paradox is, as is also well known in this forum, still subject to debates and in this sense it's a matter of opinion. For me personally there is only one satisfactory explanation, i.e., only one choice for the physical resolution of the paradox, and this is choice of the minimal statistical interpretation of the quantum state, which takes the meaning of the quantum state as only that of what's observable, and that are probabilities for the outcome of measurements of previously indetermined observables, here the polarization of the single photons.
It is also clear that the description of the measurements of the photons at the far distant places is done with "local equipment" and the only satisfactory description of the interaction of the single photons with the polarizing beam splitter and the detector at the corresponding place, here is relativistic local quantum field and thus by construction there cannot be any faster-than-light influences of one measurement on the other measurement.
So what the state of the two photons before the measurement simply says is that both observers measure completely unpolarized single photons, but there is the 100% correlation between the measured polarizations, because this correlation is already due to the preparation in this polarization-entangled state, and the reason for being in this state is simply the conservation of the total angular momentum. I.e., the quantum state describes the preparation of the two-photon system before any measurements were done, and this preparation procedure implies both, (a) the total indeterminism of the polarization of the single photon and (b) the 100% correlation between the outcome of the polarization measurements at far distant places.
No FTL communication is thus needed to explain this 100% correlation. However, if Alice and Bob want to verify that these correlations are really there, they have to exchange the outcome of measurements. E.g., for each photon Alice measures she sends a message to Bob about her outcome, and he can compare it to what he measured on his photon. However, of course, he has to wait for 2 years before Alice's messages reaches him. So there's no faster-than-light communication possible with this entanglement, because neither can Alice predetermine in any way the outcome of her measurement, because what she observes is just an ideally unpolarized photon, i.e., she cannot provide Bob with a predetermined plan for communication via the outcome of her photon-polarization measurement and in this way send instantaneous messages. All Alice and Bob can do is to "a posteriori" confirm the predicted 100% correlation between the outcome of their polarization measurement, for which they need at least two years to communicate about.
With the minimal statistical interpretation there are never any contradictions between relativistic (Einstein) causality and relativistic quantum field theory, because relativistic quantum field theory is constructed in such a way that it can never violate this causality, because it's assuming that there are only local interactions (which is the very reason already for classical physics to introduce the concept of fields instead of actions at a distance, here partiularly the electromagnetic field for electromagnetic interactions) and that the microcausality constraint holds, according to which any local observable operator commutes with the Hamilton density operator for space-like separated arguments.
With that statement you can move the entire thread to the quantum interpretation section and debate over the philosophical implications ;-)).
