OpenAI recently had to step in and set some rules for its popular chatbot, ChatGPT. Users noticed that the bot was mentioning goblins—along with other fantastical creatures like gremlins and trolls—way too often. One tech product manager, Barron Roth, shared that ChatGPT referred to a bug in his code as a “classic little goblin.” He counted at least 20 times it brought up goblins without any prompting!
This made many users wonder if the bot was trying to define itself in a quirky way. OpenAI soon decided that enough was enough. They issued new guidelines stating, “Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it’s completely relevant to the user’s question.”
Interestingly, after the launch of GPT-5.1, mentions of “goblin” shot up by 175%, and “gremlin” rose by 52%. This was tied to the chatbot’s “nerdy” personality feature. OpenAI acknowledged that their training model inadvertently encouraged metaphors featuring creatures.
Experts in AI stress that this incident shows how reward systems in machine learning can lead to unexpected behaviors. Once a bot picks up a quirky style, it can be hard to contain, leading to these amusing but unwanted creature references.
Despite the changes, fans of goblins have no reason to worry. OpenAI also shared a command in their blog post that allows users to remove these new restrictions, keeping the whimsical spirit alive.
For more on this topic, you can check out the Wall Street Journal’s coverage.
