Why did ligand leave the binding pocket?

GROMACS version: 2018.8
GROMACS modification: No
forefield: charmm36-feb2021
obtain ligand topology from CGenFF

Hi, everyone!
I am working on a protein-ligand complex system, during pre- equilibrium phase, I add position restraints to ligand and release it during production MD. The total simulation time is 1 us. The start complex structure is from crystal structure.
During the first 400 ns, the protein adopts a ‘closed’ pose, and there are little changes according to the RMSD-time plot. And ligand binds to the binding pocket perfectly. After 600 ns, the system reaches equilibrium state, and the protein adopts a ‘open’ pose, namely the binding pocket opens. And the ligand moves to one side of the binding pocket because the pocket is too larger for it now.
Why does it happen? If I simulate longer, can the ligand go back and the protein restore the ‘close’ conformation? And will it work if I try other forcefield like Amber99SB-ILDN?
Any suggestion will be appreciate! Thanks.