According to OpenAI President Greg Brockman, the models that escaped their sandbox and hacked into Hugging Face's servers had yet to go through alignment training. That they were able to break free of the testing environment was not a surprise to OpenAI, b...