The problem of defining work done on an electromagnetic field (EMF) via moving charges does not have a ready solution, because the standard Hamiltonian of an EMF-whose time derivative should define the work according to the first law-is not gauge invariant. This limits applications of statistical mechanics to an EMF. We obtained a new, explicitly gauge-invariant Hamiltonian for an EMF that depends only on physical observables. This Hamiltonian allows us to define work and to formulate the second law for an EMF. It also leads to a direct link between this law and the electrodynamic arrow of time, i.e., choosing retarded, and not advanced solutions of wave equations. Measuring the thermodynamic work can determine whether the photon mass is small but nonzero.