Yes, this is a problem that happens with a lot of detector types as one tries to push them to detect lower energies.
In this specific case, light is captured by a retinal molecule that is held within an opsin protei and this excess energy allows it to twist. The opsin protein envelope tunes the environment around the retinal molecule such that it absorbs photons of specific colors. To tune these to lower energies means altering the energy landscape in a way that makes the twisting of the retinal molecule require less energy. Temperature is a measure of the kinetic energy of molecules, and some molecules at room temperature can move with kinetic energies that match the energy of infra-red and sometimes near-infra red photons. So, from time to time, molecules with enough kinetic energy may collide against the protein with enough force to induce the twist without light. Absorbing near-IR light would lead to an increase in thermal noise, because the opsin becomes activated by a collisions more often as opposed to light.
Here is an image showing the retinal twisting and the energy landscape. The 'hv' arrows indicate light absorption, and the opsin's structure alters the energy curves.
The core-shell upconversion nanoparticles are special in that the photoactive region in the core is protected from the environment by an inert transparent shell. Light can pass through the shell and create localized electronic excitations within the core, while it is much more difficult to transfer the energy from a molecular collision against the outer layer into a localized excitation in the core. This shell is a strong dampener that protects the core from external influence.
Here is an example image of the core-shell structure, for reference:
