Worried about hackers employing LLMs to write powerful malware? Using targeted reinforcement learning (RL) to train open source models in specific tasks has yielded the capability to do just that.