OpenAI has lifted the lid (just a crack) on its safety-testing processes. It has put out two papers describing how it stress-tests its powerful large language models to try to identify potential ...