Comment by EagnaIonat

A lot of the models in Ollama you can already easily bypass safe guards without having to retrain. OpenAI's open source models can be bypassed just by disabling thinking.