Comment by willfiveash

This abliteration attack seems worrying: https://projectdiscovery.io/research/how-abliterated-models-... .