This abliteration attack seems worrying: https://projectdiscovery.io/research/how-abliterated-models-... .
You should not trust any LLM, regardless if it was abliterated or it comes from its original creator.
You should check and control what the tools invoked by an LLM may do.