They all hired the same startup security firm Irregular to try their internal unrestricted models, which is weird especially for Google to do. At OpenAI, they didn't stop the test even after it was apparent the bots had broken out. That's all I got.