Originally published at: https://tidbits.com/2026/05/05/chatgpts-goblin-obsession-evades-openais-fixes/
At Ars Technica, Kyle Orland writes:
The system prompt for OpenAI’s Codex CLI contains a perplexing and repeated warning for the most recent GPT model to “never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.”
The explicit operational warning was made public last week as part of the latest open source code for Codex CLI that OpenAI posted on GitHub. The prohibition is repeated twice in a 3,500-plus word set of “base instructions” for the recently released GPT-5.5, alongside more anodyne reminders not to “use emojis or em dashes unless explicitly instructed” and to “never use destructive commands like ‘git reset –hard’ or ‘git checkout –’ unless the user has clearly asked for that operation.”
I read this article and OpenAI’s explanatory blog post a few days ago and chalked it up to “yet another wacky AI oddity.” (The goblin behavior apparently spread from “Nerdy” personality training to the broader model through reinforcement learning feedback loops.) But despite—or perhaps because of?—the attention and OpenAI’s system prompt revisions, I was amused to see ChatGPT suddenly start sprinkling goblin references into a lengthy conversation about a presentation I’m building, with inscrutable phrases such as:
- The shiny box is doing goblin math behind a receptionist voice.
- One creates mild embarrassment; the other may summon the filesystem goblin.
- That’s the little goblin gearbox inside the velvet box.
- The spreadsheet goblin is still a goblin.
- That takes longer than a theme swap, but it avoids the little formatting goblins.
- PowerPoint-to-Keynote conversion is not magic, just a tidy little file-format goblin wearing a necktie.
- The reason is the same file-format goblin we ran into with Keynote
- Tiny ribbon goblin, wearing a Microsoft badge.
Here’s hoping that OpenAI can exorcise its goblins and other conversational tics soon. Until then, I’ve updated the first line in my ChatGPT custom instructions to:
Write in a direct, factual, concise style. Do not use filler phrases, theatrical honesty formulas, or self-announcing tone markers. State information plainly without prefacing it. Avoid motivational rhetoric, hedging clichés, or conversational padding. When providing analysis, speak in straightforward declarative sentences and focus solely on substance. And don’t mention goblins!
