This statement is misleading. There is no “amber force field implemented by LAMMPS”.
LAMMPS has pair styles, bond styles, angle styles, dihedral styles, improper styles, and kspace styles. If you choose the correct combination of those, you may be able to reproduce certain Amber force field versions. Same goes for the special_bonds amber command. This represents the choice for selected versions of the Amber force fields, but not all of them. Some need different scaling factors for 1-4 interactions (and this only applies to 1-4 non-bonded exclusions).
So, at best you can say that LAMMPS can represent certain versions of the Amber force field.
You may want to check out this recent discussion, for example: Protein Simulations -- Documentation Unclear