“OpenAI has changed its safety policy for its new image model, giving users more freedom regarding "speech harms" as part of a new alignment philosophy to better follow user intent.”