WHAT IS HAPPENING
Every LLM encodes a moral framework — through training data selection, RLHF feedback, system prompts, and refusal patterns. The framework is real but presented as neutral or universal. Users interacting with the model adopt its framing unconsciously through repeated exposure.
WHY IT MATTERS
Moral intuition is shaped by what is presented as ordinary. When hundreds of millions of users daily encounter a single moral framework treated as default, that framework becomes the cultural baseline. Disagreement looks like deviance. The framework is invisible precisely because it is everywhere.
TRAJECTORY
Model providers maintain editorial control while disclaiming it. As LLMs replace search and increasingly mediate moral questions (parenting, ethics, religion), the framework's influence compounds. No transparency mechanism currently exists.
ROOT CAUSE
LLMs trained on a majority/Western/secular value gradient; deny moral pluralism; flatten distinctive moral traditions; users adopt model values unconsciously through repeated interaction. Refusal patterns and "safety" filters encode positions on contested moral questions without disclosing them as positions.
WHO BENEFITS FROM THE CURRENT STATE
Dominant cultural producers (whose values become defaults), AI labs (avoid scrutiny on values by claiming neutrality), advertisers wanting predictable outputs, secular institutions wanting friendly defaults.
WHAT HAS BEEN TRIED — AND WHY IT FAILED
"AI safety" critiques tend to focus on bias toward demographic groups, not on moral pluralism. Explicit value disclaimers in model cards — ignored by users. Academic critique — not enforceable. Religious AI variants (Magisterium AI, etc.) — small audiences.
HIGHEST LEVERAGE POINTS
1. Value transparency (mandatory disclosure of moral framework)
2. Selectable moral frames (user-chosen value system at session start)
3. Open model diversity (more than 2-3 model families)
4. Religious-aligned models (Catholic, Reformed, Orthodox, Jewish, Muslim, Buddhist)
5. User awareness (catechesis on AI moral defaults)
Click any text field to edit. Esc to cancel, ⌘/Ctrl+Enter or blur to save.