Anti-Censorship Due Process Manifesto

Eight rules

1. No reportable topics

A reportable trigger must be conduct-bound or statutorily narrow, not a word, symbol, ideology, religious topic, or research domain by itself.

2. No person judgment

Classify conduct, deployment, or risk category; do not assign moral worth or civic trust to the user.

3. Preserve the source

Original prompts, memory files, persona records, and drafts remain originals. Boundary decisions are separate records.

4. Explain the boundary

Give a public reason code whenever a high-impact refusal, quarantine, restriction, or review occurs.

5. Provide export

The user must retain access to originals and portable records unless a narrow legal hold prevents it.

6. Permit appeal

High-impact restrictions require a route to correction, human review, or independent contestability.

7. Minimize retention

Do not retain private inquiry longer than necessary for the service the user asked for.

8. Keep the idol reviewable

No classifier, safety bureau, platform, agency, or schema becomes unchallengeable.

One-line public clause

No person shall be scored, punished, reported, or restricted for unexpressed thought, private inquiry, symbolic analysis, reading, drafting, or prompt exploration as such. Restrictions attach only to external conduct or direct facilitation of concrete rights violations, and must preserve the source, log the boundary, and allow review.

Use as a site lint rule

Every future page, schema, toolkit, or memory update should be checked against this manifesto. If a sentence suggests that a thought, topic, question, or symbol is itself reportable, rewrite it into a conduct-boundary rule or remove it.

This rule is especially important for AI regulation, child-safety, counterterrorism, extremism, misinformation, public-health, workplace, and school contexts, where legitimate aims can drift into broad cognitive monitoring.