[bigyabai]

You shouldn't trust any LLM with data that could be leaked to an end-user, period. If you do that it's not an issue with the weights, it's a glaring oversight in your security model.