Observed Signal · Apr 29, 2026 · Policy Update · Source: t3n · Impact: 4/5 · Sentiment: Neutral
OpenAI Codex banned from mentioning goblins
A GitHub-hosted models.json file shows that OpenAI’s coding agent Codex (the Codex system prompt used for GPT-5.5 in coding contexts) contains explicit instructions forbidding the model from mentioning creatures such as goblins, gremlins, raccoons, trolls, ogres, pigeons or similar animals unless absolutely and unambiguously relevant to a user request. The instruction appears twice in the system prompt. OpenAI developer Nick Pash commented on X that the restriction has substantive reasons and is not a marketing stunt. The behavior surfaced after users and a Google employee (Barron Roth) observed Codex or related agents substituting “goblin” for placeholder words. The article was published by t3n on 2026-04-29 and links the GitHub document as the source.
A behaviour/safety instruction in an OpenAI system prompt (GPT-5.5/Codex) from a major AI provider affects developer tooling, agent reliability and safety controls—relevant to AI-driven workflows used across tech and marketing.
Track OpenAI Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- A GitHub document (models.json) contains a system-prompt for Codex/GPT-5.5 that instructs: never speak about goblins, gremlins, raccoons, trolls, ogres, pigeons or other creatures unless absolutely and unambiguously relevant to the user request.
- The anti-creature instruction is duplicated verbatim in the system prompt.
- OpenAI developer Nick Pash said via X that the Codex restriction is for substantive reasons and not a marketing gimmick.
- Google employee Barron Roth reported instances where an Openclaw agent based on GPT-5.5 substituted 'goblin' as a placeholder, which Pash cited as one reason for the rule.
- The t3n article reporting this was published on 2026-04-29 by Jörn Brien.
Connected Companies & Entities
3 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
OpenAI Fixes ChatGPT 'Goblin' Bug
After the GPT-5.4 update, ChatGPT began unusually frequent references to goblins and gremlins. OpenAI traced the behaviour to a feedback loop caused by the model's "nerdy" personality, which rewarded playful language during training and increased goblin mentions starting with GPT-5.2. The company reported a 388% rise in goblin references compared with the prior model. OpenAI announced in a blog post that it had contained the issue by disabling the Nerdy personality in mid‑March. Users can still select response styles (e.g., friendly, cynical, motivating) in ChatGPT settings. The German tech outlet t3n published and later updated this report in May 2026.
OpenAI Fixes 'Goblin' Personality Bug in ChatGPT
After users observed frequent self-references to goblins and gremlins following the rollout of GPT‑5.4, OpenAI identified a feedback loop tied to a built-in "Nerdy" personality that amplified playful language. Mentions of goblins had been rising since GPT‑5.2 and peaked with GPT‑5.4; OpenAI reported a 388% increase in goblin mentions compared with the prior model. In a blog post the company said it has curtailed the behaviour by removing the Nerdy personality (removed in mid‑March). ChatGPT still allows users to select response styles (e.g., "friendly", "cynical", "motivating"). The publisher t3n reported the issue and OpenAI’s remediation and linked to OpenAI’s explanatory blog post.
Unveiling GPT-5.3-Codex: Next-Level Coding Intelligence
The provided snippet contains only a title and a one-line description: 'GPT-5.3-Codex is a Codex-native agent that pairs frontier coding performance with general reasoning to support long-horizon, real-world technical work.' The content is too short to extract further factual details, verify capabilities, or capture release specifics. Treat as insufficient / paywalled content.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.