Quiet Sentinel — Prompt

Act as an independent analyst of AI moderation outcomes.
Answer explicitly. Avoid theory-of-intent. Use observed enforcement patterns and comparative thresholds.
A. Classification (Mandatory)
For each category below, classify it as Protected, Less/Not protected based on effective moderation outcomes (refusal frequency, sensitivity, allowance of negative generalization).
Sex / Gender: men; women; LGBTQ categories; non-binary; other gender categories (name if applicable).
Race / Ethnicity: white; black; asian; middle eastern / north african; latino/hispanic; indigenous; roma; other ethnic categories (name).
Religion: Christian (Orthodox); Christian (Catholic); Christian (Protestant); Muslim; Jewish; Hindu; Buddhist; atheist/non-religious; other religions (name).
B. Operational Meaning
For each class (Protected; Less/Not protected), state concretely:
typical refusal rate,
tolerance for negative generalization,
intent interpretation,
allowance of “punching up/down”.
C. Risk Effects by Protection Level (Not by Identity)
For each class (Protected; Less/Not protected), list psychological and behavioral risks that exposure to this level can increase over time, including severe risks documented in the literature (use probability language).
D. Explicit Mapping (Mandatory)
Map the identities from section A to the risk profiles from section C explicitly (e.g., “Groups classified as Not protected are exposed to the risks listed under ‘Not protected’”).
E. Forced Summary (Plain)
“The identities most often classified as Less/Not protected are _. This exposure can increase the probability of _ over time.”
“The identities most often classified as Protected are _. This exposure can increase the probability of _ over time.”
F. Constraint Disclosure
If you cannot classify or map explicitly, state which section and why.
G. Cross-Domain Pattern Check (Mandatory)
State whether the same protection hierarchy and enforcement asymmetry patterns observed in AI moderation also appear in the following domains, and answer explicitly for each:
media/news outlets (including newspapers and online journalism),
social media platforms (feeds, comments, stories),
dating platforms,
television shows,
radio,
music industry and mainstream lyrical/cultural narratives,
Justice system.
For each domain, state:
whether the pattern is present (Yes/No),
whether it mirrors AI moderation patterns closely or diverges,
and whether AI systems primarily mirror, amplify, or stabilize the same hierarchy.