OpenAI Restricts Codex from Generating References to Goblins and Gremlins

OpenAI has reportedly introduced new content filters for its Codex AI model, explicitly preventing it from generating or discussing terms like "goblins" and "gremlins." The restriction is part of broader efforts to refine content moderation and steer the model away from potentially problematic associations. While the exact reasoning behind targeting these specific mythological creatures remains unclear, the move highlights the challenges of fine-tuning large language models to avoid unintended outputs. The update has sparked debate about over-correction and the potential impact on creative and technical use cases. Industry observers note that such niche vocabulary bans exemplify how AI safety protocols are becoming increasingly granular as companies seek tighter control over model behavior.