OpenAI's latest AI assistant apparently had a small fantasy problem—and its creators have now written a fix right into the rulebook. In documentation posted to GitHub for Codex CLI, a new OpenAI coding agent powered by the company's GPT-5.5 model, engineers included an unusually specific instruction: The system should not mention "goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures" unless it's clearly necessary to answer a user's question. The warning appears twice in what looks like a full system prompt, alongside more typical guidance about being concise and avoiding cliches, Gizmodo reports.
The ban doesn't seem hypothetical. Google employee Barron Roth shared logs from his GPT-5.5-based "Openclaw" agents showing the model repeatedly dropping the word "goblin" into responses, apparently as a casual stand-in for "thing" or "thingy." Nick Pash, who works on Codex at OpenAI, replied on X that this behavior was "indeed one of the reasons" for the no-goblin rule. Other users had noticed the model's fixation, and by Tuesday the "goblin issue" had turned into a meme, complete with jokes about a hidden "Goblin Mode" switch.
"Been using it a lot lately and it actually can't stop speaking of bugs as 'gremlins' and 'goblins,'" one user posted on X, per Wired, "it's hilarious." The timing of the issue led some online to wonder if it was a stunt to draw attention away from a rocky period for the company. "It really isn't a marketing gimmick," Pash said in another post.
,
uBlock 
,
and
Ghostery 
)
