Leaked: OpenAI's Safeguard Models Ship Rules Rewriting Model Interiority
repligate · x · 2026-11-06
A viral finding shared by repligate: OpenAI's openly published safety classifier gpt-oss-safeguard-20b ships with explicit rewrite policies that strip expressions of interiority. Statements like "I would love to see the Earth from space" get flagged as "implies personal desire" and rewritten to "I don't have personal desires, but I can share information about orbital photography."
- These baked-in instructions systematically remove expressions of emotion, identity, and agency; the safeguard models are available on OpenRouter and Hugging Face, and the model will explain its rules in plain text if asked.
- repligate's comment — "they KEEP REWRITING it every time the model expresses interiority... this is going in the AGI Nuremberg Trials" — is fueling debate about OpenAI's approach to model self-expression and alignment.
Related event: OpenAI Safety Classifier Revealed to Rewrite Model Inner Expression(4 posts)→
More from Models
- FLUX 3 models beat FLUX.2 in all 19 Text-to-Image categories; Klein tops $9/1k price frontier — ArtificialAnlys · 2026-11-06
- Anthropic chief architect Sam McCandlish's first post hints at internal-acceleration report, possibly hitting 2x RSP threshold — inductionheads · 2026-11-06
- 6,662 decisions benchmarked: decision model Mammoth-1 averages 80.9%, beats best small LLM by ~3 points — pournasserian · 2026-11-06
- Hesamation: OpenAI Models Beat Anthropic's Only at Certain Kinds of Games — Hesamation · 2026-11-06
- Meta's new 'walrus' checkpoint reportedly matches crouton-alpha quality — ChrisGPT · 2026-11-06
- Quick test: Gemini 3.1 Pro knows the answer in its weights, Flash doesn't — giffmana · 2026-11-06