CHARMM parameterization destroys ligand structure

GROMACS version: 2023.3
GROMACS modification: No


I have successfully parametrized ligands for a Protein-Ligand MD simulation with CGenFF before. However, I noticed that with a specific ligand the cgenff_charmm2gmx script produces a LIGANDNAME_ini.pdb file in which the ligands structure is completely lost in comparison with the input mol2 file.
To show what I mean I have uploaded a pictures of the structure after parametrization. As visible this is not a normal small molecule structure but rather a mess of atoms and bonds.

I can also give the mol2 file of the ligand (but apparently its not allowed to upload this here).

The to be parametrized ligand was one that came out of a ML generative model and for which I then generated a 3D structure in rdkit. Is there maybe some kind of bias in the cgenff_chamm2gmx script so that “unrealistic” molecules do not get parametrized properly ? In this case I think some kind of warning or failing behavior would be expected.
In any case I hope someone can help with this.