OpenAI says it has made several changes to its safety practices following the Hugging Face breach and has paused two weeks of deployment-focused RL training (Ina Fried/Axios)