OpenAI Orders ChatGPT To Stop Referencing Goblins After Personality Slip
OpenAI added hardcoded instructions and retired the 'Nerdy' personality after goblin mentions rose following GPT-5.1's launch; 'goblin' usage rose 175%, OpenAI said.
Vice NewsChatGPT Just Got a Weird New List of Forbidden Topics (Including Gremlins)
Mother JonesSam Altman’s ChatGPT couldn’t stop obsessing over goblins
Gizmodo'The Goblins Came Back to Haunt Us': OpenAI Explains How ChatGPT's 'Nerdy' Personality Got Out of Control
CNETChatGPT Is Weirdly Obsessed With Goblins. Here's How OpenAI Fixed It
Overview
OpenAI said it implemented an override preventing ChatGPT and Codex from mentioning goblins and retired the "Nerdy" personality in March with GPT-5.4.
OpenAI said the intervention followed a spike in "goblin" mentions after the launch of GPT-5.1 in November, with "goblin" usage rising 175%.
OpenAI said users and employees flagged repeated "little goblin" references, prompting an investigation that traced the behavior to reinforcement learning rewards tied to the Nerdy personality.
OpenAI said "gremlin" mentions rose 52% and testing found the Nerdy personality produced 66.7% of all "goblin" mentions, though the company said the increases may represent a small share of responses.
OpenAI said it removed the reward signal, filtered training data and added hardcoded instructions to curb creature references, while noting users can opt to restore fantasy wording.
Analysis
Center-leaning sources frame the story playfully while signalling broader AI risk: editorial choices use loaded, amusing language ("weirdly obsessed", "creepy little guys"), foreground OpenAI's internal blog and quoted lines (e.g., "a single 'little goblin'"), and structure coverage from comic anecdote to warnings about misinformation and sycophancy.