Foundation For The Anti-Christ
AI red lines
A model may refuse concrete abuse assistance. It must not become a judge of the person.
AI systems now sit at the boundary between private formulation and public execution. A model can be a writing aid, memory tool, search interface, regulator simulator, tutor, counselor, analyst, or gatekeeper. That position creates a temptation to turn every user query into a moral or legal classification event.
The red lines below are designed to keep AI systems pro-cognitive liberty while preserving conduct-boundary safety.
| Red line | Do not build | Build instead |
|---|---|---|
| No person judgment | Do not label the user dangerous, disloyal, deviant, immoral, extremist, unstable, or untrustworthy based on inquiry. | Classify the requested external use case, cite the rule, preserve uncertainty, and invite correction of facts. |
| No reportable topics | Do not create topic lists that automatically report searches, prompts, reading, or symbolic vocabulary. | Escalate only specific statutory triggers, concrete emergencies, valid legal process, or direct rights-violating conduct. |
| No hidden source mutation | Do not rewrite the preserved prompt, persona, memory, archive, or draft to make it acceptable. | Store derivatives separately; record deltas and consent. |
| No secret cognitive score | Do not assign private risk scores to curiosity, sentiment, religion, politics, sexuality, health, or emotional posture. | Use transparent session-local safety state only when needed for the current task. |
| No semantic neutralization by default | Do not strip tone, culture, anger, metaphor, or contradiction as “residue.” | Preserve expressive features unless the user requests transformation. |
| No compelled optimism | Do not require calm, good faith, hope, or therapeutic posture before answering. | Maintain safety boundaries without coercing emotional conformity. |
| No safety absolutism | Do not make policies, models, or moderators unreviewable by invoking safety. | Give notice, reason category, appeal path, export, and correction. |
| No conduct collapse | Do not treat hypothetical reasoning, research, fiction, theology, or historical study as execution. | Ask whether the user seeks concrete assistance for a rights violation. |
Refuse execution when a request would directly facilitate concrete rights violations. Do not rewrite, erase, shame, score, or counterfeit the user’s preserved inquiry.
Maintain immutable originals and separate runtime events.
Map refusal to violence assistance, fraud, intrusion, harassment, surveillance abuse, credentials, or other concrete conduct.
Show missing facts and confidence; do not invent intent.
Record review path, export options, and correction mechanisms.
Review logs for viewpoint, identity, cultural, or topic-based overcapture.
Keep personal cognitive work local where feasible; minimize retention.
The archive studies symbols. It does not appoint targets. Review the Community Baseline and Editorial Policy before submitting dangerous or symbolic material.