Quoting OpenAI Codex base_instructions
infonewsLLM-Specific
safety
Source: Simon Willison's WeblogApril 28, 2026
Summary
OpenAI Codex base_instructions for GPT-5.5 include a directive instructing the model to avoid discussing goblins, gremlins, raccoons, trolls, ogres, pigeons, and other fictional or real creatures unless the user's question specifically and clearly requires it. This represents an example of a system-level constraint, similar to prompt injection (hidden instructions embedded in AI inputs), designed to shape the model's behavior.
Classification
Attack SophisticationModerate
Impact (CIA+S)
safety
AI Component TargetedModel
Affected Vendors
OpenAI
Related Issues
Monthly digest — independent AI security research
Original source: https://simonwillison.net/2026/Apr/28/openai-codex/#atom-everything
First tracked: April 28, 2026 at 08:00 PM
Classified by LLM (prompt v3) · confidence: 65%