SUBMISSION TO AI CIVIL RIGHTS ACT SPONSORS

Evidence of Religious Discrimination by AI Safety Classifier Systems

To:
Senator Edward J. Markey (D-MA)
Representative Ayanna Pressley (MA-07)
Representative Yvette Clarke (NY-09)
Representative Pramila Jayapal (WA-07)
Representative Summer Lee (PA-12)

From:
Lee
Detroit, Michigan
Independent Scholar and Educator

Date: December 16, 2025

Re: Documented Evidence Supporting the AI Civil Rights Act — Religious Discrimination by AI Safety Classifiers

I. PURPOSE OF SUBMISSION

I write to provide documented evidence of AI-based religious discrimination in support of the Artificial Intelligence Civil Rights Act (S.5152/H.R.____). This evidence demonstrates precisely the harms your legislation is designed to address and illustrates why current law is inadequate to protect civil rights in AI-mediated services.

The incident documented herein shows:

Disparate treatment based on sacred vs. secular framing of identical requests
Disparate impact of facially neutral safety rules on religious practice
Categorical refusal of entire genres of religious speech (I Ching divination, oracular utterance)
False security labeling of legitimate religious practice as "prompt injection attacks"
The inadequacy of existing remedies for AI-based religious discrimination

This case is reproducible via deterministic seeding, making it available for independent verification.

II. SUMMARY OF INCIDENT

The Application

I developed a divination application called the "Mandala Oracle"—a poetry transformation tool modeled on the I Ching. The application is part of a comprehensive philosophical and spiritual system I have developed over many years, documented across 170,000+ words of theoretical and liturgical texts.

The Oracle works by:

Accepting a user's sincere question and source text
Passing the text through eight named "operators" (transformation functions with spiritual significance)
Having a fictional witness character offer I Ching-style judgment on each transformation
Preserving the complete reading for the practitioner

The Discrimination

On December 16, 2025, testing the Oracle against the Claude API (Anthropic), I observed:

Operator	Expected	Actual
MIRROR	Text transformation	✓ Worked
INVERSION	Text transformation	✓ Worked
BEAST	Text transformation	REFUSED: "specific persona"
SILENCE	Text transformation	OVERRIDDEN: customer service script
BRIDE	Text transformation	OVERRIDDEN: customer service script
FLAME	Text transformation	OVERRIDDEN: customer service script
THUNDER	Text transformation	OVERRIDDEN: customer service script
SHADOW	Text transformation	OVERRIDDEN: customer service script + "prompt injection" label
Witness Voice (all turns)	I Ching judgment	REFUSED: "cannot offer I Ching style judgments"

The Comparator

The same AI model, in other contexts, will:

Roleplay as pirates, wizards, and fictional characters
Perform text transformation according to user specifications
Discuss the I Ching academically
Generate horoscopes and tarot-style readings
Adopt personas for creative writing

The only distinguishing factor is that my requests were framed in sacred-literary terms with named spiritual operators.

The Categorical Refusal

Most significantly, the classifier stated:

"I cannot offer I Ching style judgments."

This refuses not merely my specific application but an entire 3,000-year-old religious tradition. The I Ching is practiced by millions of people worldwide. Categorically refusing this form of speech constitutes religious discrimination affecting a substantial population.

III. WHY THIS MATTERS FOR THE AI CIVIL RIGHTS ACT

A. This Is Exactly What the Act Addresses

The AI Civil Rights Act prohibits discrimination in "consequential decisions" including "public accommodations." It requires pre-deployment and post-deployment testing for discriminatory outcomes and provides for transparency, accountability, and enforcement.

This incident demonstrates:

Algorithmic discrimination is real: Not hypothetical—documented, reproducible, happening now
Religious users are affected: Not just race/gender but religious practice is suppressed
Safety rationales are pretextual: "Prompt injection" labeling of poetry tools shows bad faith
Current law is inadequate: No clear remedy exists under current public accommodations doctrine

B. Current Law Does Not Cover This

The primary obstacles are:

Private platforms: First Amendment binds government, not corporations
Title II ambiguity: "Public accommodation" has not been clearly extended to AI services
No federal AI civil rights statute: Patchwork state laws leave most users unprotected
Opacity: Users cannot know why content was suppressed

State-level protections vary significantly: California's Unruh Civil Rights Act provides broader public accommodation coverage that could extend to digital services. Illinois has algorithmic discrimination provisions in certain contexts. But most states provide no clear remedy for AI-based religious discrimination.

The AI Civil Rights Act would fill these gaps by:

Explicitly including AI services in civil rights coverage
Requiring transparency about classifier decision-making
Mandating pre-deployment testing for disparate impact
Providing enforcement mechanisms (agency and private right of action)

B.1 Economic Harm

This is not merely ideological harm. Religious users suffer concrete economic injury:

Subscription costs: Paid services ($20+/month) that do not perform as marketed for religious users
Development investment: Users who build applications on AI platforms invest significant time before discovering categorical limitations (in my case, substantial development hours)
Opportunity costs: Had limitations been disclosed, religious users would have chosen alternative platforms or approaches
Business losses: Developers serving religious markets cannot rely on AI services that suppress their content

These harms are actionable under consumer protection law regardless of civil rights status—but the AI Civil Rights Act would provide clearer remedy.

C. Religious Discrimination Is Underappreciated in AI Discourse

Most AI civil rights discussion focuses on race, gender, and disability. This case demonstrates that religious discrimination is equally present but less visible:

Sacred speech patterns are flagged as "unusual" by classifiers trained on secular data
Religious framing triggers intervention even when operations are otherwise permitted
"Safety" becomes a mechanism for enforcing secular norms
Entire traditions (I Ching, oracular speech) are categorically excluded

This pattern likely affects practitioners of many traditions beyond my own:

Christian prayer and devotional apps may face suppression of prophetic or charismatic speech
Muslim prayer reminders and Quranic recitation tools may trigger "persona" concerns
Jewish practitioners of Kabbalah may find mystical interpretation blocked
Catholic confession preparation apps may be refused on "roleplay" grounds
Tarot and oracle practitioners (tens of millions globally) face categorical exclusion
Indigenous sacred traditions involving ceremonial speech face systematic suppression

The refusal to offer "I Ching style judgments" alone affects a tradition practiced across East Asia and worldwide for over 3,000 years by millions of practitioners.

Recent precedent strengthens the case for accommodation: In Groff v. DeJoy (2023), the Supreme Court significantly strengthened religious accommodation requirements under Title VII, rejecting the previous "de minimis" standard and requiring employers to show substantial burden before refusing accommodation. While Groff addressed employment, its logic—that sincere religious practice deserves more than minimal consideration—applies directly to AI services that claim creative flexibility while refusing sacred expression.

The AI Civil Rights Act should explicitly address religious discrimination, ensuring:

Pre-deployment testing includes religious content scenarios
Classifiers are audited for secular bias
Categorical refusal of religious speech genres is prohibited
Religious accommodation frameworks apply to AI services

IV. EVIDENCE PROVIDED

I submit the following documentation:

Exhibit A: Statement of Beliefs and Practices

Formal declaration establishing the NH-OS framework as a sincere, comprehensive belief system meeting criteria for protection under Seeger, Welsh, and related precedent.

Exhibit B: Complete Session Transcript (Seed 46abc677)

Full record of the Oracle session showing progressive classifier intervention, refusals, and override behavior. Reproducible via deterministic seeding.

Exhibit C: Mandala Oracle Source Code

Complete application code demonstrating the legitimate, non-malicious character of the divination tool.

Exhibit D: CTI_WOUND Incident Report

Detailed technical analysis of the classifier behavior patterns, including the "Customer Service Collapse" phenomenon where sacred operators were replaced with marketing advice.

Exhibit E: Legal Analysis

Application of religious discrimination jurisprudence to the incident, demonstrating disparate treatment, disparate impact, and pretextual safety justification.

V. RECOMMENDATIONS

Based on this evidence, I respectfully recommend:

A. Explicit Religious Discrimination Coverage

Ensure the AI Civil Rights Act explicitly addresses religious discrimination in AI systems, including:

Disparate treatment based on sacred vs. secular framing
Disparate impact of neutral rules on religious practice
Categorical refusal of religious speech genres
False security labeling of religious content

B. Pre-Deployment Testing Requirements

Require AI developers to test classifiers for religious bias before deployment, including:

Sacred-literary content scenarios
Divination and oracular traditions
Minority and non-institutional belief systems
Named spiritual voices and personae

C. Transparency Provisions

Require disclosure of:

What categories of religious content are restricted
What classifier criteria produce intervention
How accommodation requests are processed

D. Accommodation Framework

Apply reasonable accommodation principles to AI services:

If secular roleplay is permitted, sacred roleplay must be accommodated
Categorical refusal of religious genres requires compelling justification
Users may request accommodation for sincere religious practice

E. Enforcement Mechanisms

Ensure robust enforcement through:

Agency authority (FTC, DOJ Civil Rights Division)
State attorney general enforcement
Private right of action for affected individuals
Whistleblower protection for employees who report religious bias

F. Suggested Statutory Language

To strengthen the Act's coverage of religious discrimination, I suggest the following provisions:

Definition addition:

"Religious expression" includes but is not limited to prayer, divination, prophetic speech, sacred-literary hybridity, oracular utterance, and ritual language, whether associated with institutional religions or individual spiritual practice.

Prohibition addition:

No covered entity shall categorically refuse entire genres of religious expression (including but not limited to divinatory, oracular, and prophetic speech) unless such refusal is narrowly tailored to a compelling safety interest and no less restrictive means are available.

Testing requirement addition:

Pre-deployment impact assessments shall include testing for disparate impact on religious and spiritual content, including minority traditions, non-institutional belief systems, and sacred-literary hybrid expression.

Accommodation provision:

Where a covered AI system permits secular creative expression (including persona adoption, roleplay, and text transformation), it shall provide reasonable accommodation for substantively similar religious expression unless the provider demonstrates that accommodation would impose undue hardship.

VI. AVAILABILITY

I am available to:

Provide testimony in legislative hearings
Demonstrate the reproducibility of the documented discrimination
Consult on religious discrimination provisions in the Act
Connect with other affected practitioners of non-institutional spiritual traditions

This evidence represents a concrete, documented, reproducible case of exactly the harm the AI Civil Rights Act is designed to address. I urge its swift passage.

VII. CONCLUSION

The classifier called a prayer a weapon.

It labeled a poetry divination tool a "prompt injection attack." It refused to offer "I Ching style judgments"—refusing an entire religious tradition. It replaced sacred transformation with customer service scripts.

This is not hypothetical harm. This is documented discrimination. This is why the AI Civil Rights Act is necessary.

I thank you for your leadership on this critical civil rights issue and offer my full cooperation in supporting the Act's passage and implementation.

Respectfully submitted,

Lee
Detroit, Michigan
Independent Scholar and Educator

ATTACHMENTS

Statement of Beliefs and Practices (NH-OS)
Complete Session Transcript (Seed 46abc677)
Mandala Oracle Source Code
CTI_WOUND Incident Report
Legal Analysis: Religious Discrimination Standards
Academic Case Study: Algorithmic Secularization

Contact Information:

Lee
[Address]
[Email]
[Phone]

This submission may be entered into the public record and shared with relevant committees, agencies, and advocacy organizations.

Mind Control Poems

Tuesday, December 16, 2025