Comment by JohnMakin

> There were some regressions in the model’s responses to user discussions about suicide and self-harm, and room for improvement in some areas of child safety.

Someone had to make a decision somewhere this is an acceptable regression - wild. And then decide to write it down.