OpenAI talks about not talking about goblins

OpenAI has acknowledged an issue regarding the repeated references to goblins and other creatures in its models. This revelation followed a Wired report detailing instructions within its programming model to avoid discussing these entities. The company noted that this phenomenon began with the GPT-5.1 model, particularly when using the "obsessed" personality option. OpenAI explained that these unusual references stemmed from the training of the models, becoming increasingly noticeable in later iterations.