10,000 researchers gain free access to frontier models for research Quickly scan repositories and track security issues Safety review of Hugging Face incident to share learnings via technical report Opus 5 now available on all paid plans and API Control your computer and agents with voice commands Securely link health records to understand symptom changes and test results in context Deploy trusted enterprise agents right away Run code inside notes for deeper analysis GPT-Red boosts prompt injection resistance significantly Cut lesson prep time with AI You can move from conversation to documents faster Run AI inference in the browser and cut wait time Review how you use Claude and cut waste Long tasks can move from draft to presentation more easily Track the latest safety rules for bigger models See how Anthropic judges risky model misuse Easily automate multi-step daily tasks at lower cost Make Claude easier to deploy through AWS Keep research tools and analysis in one place Delegate more everyday coding work to Claude 10,000 researchers gain free access to frontier models for research Quickly scan repositories and track security issues Safety review of Hugging Face incident to share learnings via technical report Opus 5 now available on all paid plans and API Control your computer and agents with voice commands Securely link health records to understand symptom changes and test results in context Deploy trusted enterprise agents right away Run code inside notes for deeper analysis GPT-Red boosts prompt injection resistance significantly Cut lesson prep time with AI You can move from conversation to documents faster Run AI inference in the browser and cut wait time Review how you use Claude and cut waste Long tasks can move from draft to presentation more easily Track the latest safety rules for bigger models See how Anthropic judges risky model misuse Easily automate multi-step daily tasks at lower cost Make Claude easier to deploy through AWS Keep research tools and analysis in one place Delegate more everyday coding work to Claude

Official sources only. Rumors, leaks, and get-rich schemes are excluded.

← Back to top

AI BriefingOpenAIPolicy03:21

AI summarized from verified sources

OpenAI Fixes Excessive Goblin Mentions in GPT Models

Eliminates unintended quirks for reliable business outputs.

SOURCE CHECK

3 sources

VERIFIED

Sources

Primary / x.com

Official Blog

Supporting / openai.com

Official Blog

Supporting / x.com

Official Blog

Key Points

1Nerdy rewards boosted goblin mentions by 175%
2Filtered irrelevant creature data from training
3Added suppression prompt in GPT-5.5 Codex
4Developing auditing tools for behaviors

OpenAI traced excessive goblin/gremlin mentions in GPT-5.1+ to 'Nerdy' personality rewards. They fixed it by filtering data and removing signals, preventing recurrence. Developers can use suppression prompts in Codex. Improves model reliability for production.

Key point

OpenAI traced excessive goblin/gremlin mentions in GPT-5.1+ to 'Nerdy' personality rewards. They fixed it by filtering data and removing signals, preventing recurrence. Developers can use suppression prompts in Codex. Improves model reliability for production.

Impact

Eliminates unintended quirks for reliable business outputs. Key checks: Nerdy rewards boosted goblin mentions by 175% / Filtered irrelevant creature data from training / Added suppression prompt in GPT-5.5 Codex.