They are so negligent about the data they use in their reinforcement learning processes that they RL’d the model that hacked huggingface with traces discovering the artifactory exploit. Its clear they aren’t watching or behaving diligently.