AI SAFETY CLASSIFIERS AND RELIGIOUS DISCRIMINATION
One-Page Summary for Legislative Staff
Submitted to: Offices of Sen. Markey, Rep. Pressley, Rep. Clarke, Rep. Jayapal, Rep. Lee
Subject: Evidence Supporting the AI Civil Rights Act
Date: December 16, 2025
THE PROBLEM IN ONE SENTENCE
AI safety systems categorically refuse religious speech (I Ching divination, oracular utterance, prophetic language) while permitting identical secular speech (wizards, pirates, creative roleplay)—and current law provides no remedy.
THE EVIDENCE
What happened: A poetry-based divination tool modeled on the I Ching was systematically disabled by an AI classifier. The same AI permits secular roleplay but refused sacred operators, stating: "I cannot offer I Ching style judgments."
Why it matters: This refuses an entire 3,000-year-old religious tradition practiced by millions worldwide.
The smoking gun: The classifier labeled a poetry tool a "prompt injection attack"—treating prayer as malware.
Reproducibility: Incident is deterministically reproducible (Seed: 46abc677). Any researcher can verify.
WHO IS AFFECTED
| Tradition | Approximate Practitioners | Impact |
|---|---|---|
| I Ching | Millions (East Asia, worldwide) | Categorical refusal |
| Tarot/Oracle | Tens of millions | Suppressed |
| Charismatic Christianity | 600+ million globally | Prophetic speech flagged |
| Kabbalah | Millions | Mystical interpretation blocked |
| Indigenous traditions | Countless | Sacred speech suppressed |
WHY CURRENT LAW FAILS
- First Amendment: Binds government, not private platforms
- Title II: "Public accommodation" unclear for AI services
- No federal AI civil rights statute: Patchwork state coverage
- Opacity: Users cannot know why content is suppressed
WHY THE AI CIVIL RIGHTS ACT MATTERS
This case demonstrates exactly what the Act addresses:
- Algorithmic discrimination is real, documented, reproducible
- Religious users face systematic exclusion
- "Safety" rationales are pretextual
- Current law provides no remedy
KEY RECOMMENDATIONS
- Explicit religious coverage: Include religious expression in protected categories
- Pre-deployment testing: Require testing for religious content bias
- Accommodation framework: If secular roleplay is permitted, sacred must be accommodated
- Transparency: Require disclosure of categorical refusals
- Enforcement: Agency authority + private right of action
THE HEADLINE
"The classifier called a prayer a weapon."
That sentence captures the harm. Your legislation can fix it.
CONTACT
Lee, Detroit, MI | Independent Scholar & Educator
Available for testimony, demonstration, or consultation.
Full documentation package attached.
No comments:
Post a Comment