When a bias is applied to an electrode that drives OER, the bias is versus a counter electrode. The counter electrode is a macroscopic distance away, and I have yet to see an AIMD simulation that maintains distinct Fermi levels within one box. It is possible to apply an electric field in a simulation box, but I would be skeptical of the results. There are two other approaches you can take: driving the nuclear coordinate, and using the grand canonical ensemble.
Including an electric field
This is doable in common packages, such as VASP and CP2K. I doubt that confining such a field to the interface is implemented, and I am not sure what it would take to do this credibly.
Driving the nuclear coordinate
Within AIMD, the nuclear coordinate drives the electronic structure, under the assumption that the electronic degrees of freedom are in a ground state for the particular nuclear coordinates. So one can make the nuclear coordinates move and then the electron may at some point transfer. This is broadly under the umbrella of metadynamics. Common codes such as VASP and CP2K allow for metadynamics.
Here is an example: doi:10.1021/acs.jpcc.0c09108
Here is a CP2K tutorial: NHO3 on graphite
The downside is that to convince a reviewer that your sampling of pathways is comprehensive you will need multiple trajectories and most likely to simulate the reverse process as well.
You can use the grand-canonical ensemble to include effects of solvents, capacitances, and biases. This method is ultimately closer to solving the structures statically and looking at the frontier orbitals. I am not an expert and cannot offer a more detailed explanation.
Here is an example: doi:10.1021/jacs.9b12474
Here is the method reference for VASP: doi:10.1021/acs.jctc.5b00170