Granted, the blog post is about opening the model up for red-teaming, so highlighting potential vectors for abuse is actually the desired intent.