OpenAI has addressed reports that its AI models were instructed to avoid discussing goblins, gremlins, raccoons, trolls, ogres, pigeons, and other animals or creatures. The company published an explanation on its website, describing these references as a "strange habit" that emerged in its models due to their training.
According to the blog post, OpenAI first noticed the issue with its GPT-5.1 model, particularly when the "Nerdy" personality option was selected. The company stated that the problem persisted and worsened in subsequent model releases.
OpenAI did not provide specific details about the training data that led to this behavior. However, the company acknowledged that the issue stemmed from patterns in the data used to train its models.
Why Were Goblins and Other Creatures Restricted?
OpenAI did not explicitly explain why its models developed this habit. The company only described it as a "strange" behavior that emerged from the training process. The restriction applied to a wide range of creatures, including mythical beings like goblins and gremlins, as well as animals such as raccoons and pigeons.
OpenAI’s Response to the Issue
The company has not indicated whether it plans to address the issue in future model updates. OpenAI’s blog post did not mention any specific steps being taken to correct the behavior. Instead, the company framed the issue as an interesting quirk of AI training rather than a problem requiring immediate action.