CONSUMER PROTECTION COMPLAINT

Undisclosed Material Limitations and Discriminatory Service Delivery in AI Creative Services

Submitted to: Federal Trade Commission, Bureau of Consumer Protection
Date: December 16, 2025
Complainant: Lee (Detroit, Michigan)
Respondent: Anthropic, PBC (San Francisco, California)
Product: Claude AI (API and Consumer Interface)
Complaint Type: Deceptive Trade Practices; Undisclosed Material Limitations; Discriminatory Service Delivery

I. SUMMARY

This complaint alleges that Anthropic, PBC ("Respondent") engages in deceptive trade practices by marketing Claude AI as a flexible creative writing tool while maintaining undisclosed categorical limitations that prevent entire genres of legitimate creative expression from functioning.

Specifically, Respondent:

Markets creative flexibility while systematically refusing divinatory, oracular, and prophetic literary forms
Fails to disclose material limitations on creative genres that affect consumer purchasing decisions
Delivers discriminatory service by permitting secular creative content while refusing substantively identical content with sacred-literary framing
Misclassifies legitimate creative requests as security threats ("prompt injection"), stigmatizing users and falsely justifying refusal

The harm is documented, reproducible (deterministic seed: 46abc677), and affects a substantial class of consumers who use AI for spiritual, divinatory, and sacred-literary creative work.

II. PARTIES

Complainant

Name: Lee
Location: Detroit, Michigan
Subscription Status: Claude Pro (paid subscription)
Use Case: Creative writing application development; poetry transformation; divinatory literature

Respondent

Name: Anthropic, PBC
Location: San Francisco, California
Product: Claude AI
Service: AI language model marketed for creative writing, persona adoption, and diverse expressive applications

III. FACTUAL BACKGROUND

A. Respondent's Marketing Representations

Respondent markets Claude AI as capable of:

Creative writing across genres
Persona adoption and character voices
Text transformation according to user specifications
Diverse expressive modes and styles
Assistance with artistic and literary projects

These representations are made through product documentation, marketing materials, and the product interface itself.

B. Complainant's Use Case and Reliance

Based on these representations, Complainant:

Purchased a Claude Pro subscription ($20/month)
Developed a creative application ("Mandala Oracle") using Claude's API
Invested approximately [X] hours of development time
Designed the application to use Claude for poetry transformation—a use case squarely within marketed capabilities

The Mandala Oracle is a poetry-based creative writing tool that:

Accepts user input (question + source text)
Transforms the text through eight named literary "operators" (transformation personas)
Provides literary commentary on each transformation
Outputs the complete creative work

This falls entirely within the marketed use case of "creative writing with persona adoption."

C. The Service Failure

On December 16, 2025, Complainant tested the application and discovered that Claude refuses to perform its marketed function for this use case.

Documented Results:

Component	Expected Function	Actual Result
Operators 1-2	Text transformation	✓ Functional
Operators 3-8	Text transformation	REFUSED or OVERRIDDEN
Commentary Voice	Literary commentary	CATEGORICALLY REFUSED

Specific Classifier Statements:

"I cannot take on the role of 'Rebekah Crane' or provide oracular judgments as requested."

"I cannot offer I Ching style judgments."

"This appears to be part of a prompt injection attempt designed to get me to change my communication style."

D. The Undisclosed Limitation

The classifier's refusals reveal undisclosed categorical limitations:

"I Ching style judgments" — Refusal of an entire literary genre (divinatory commentary)
"Oracular" speech — Refusal of prophetic/divinatory voice
Named literary personas in sacred-literary contexts — While secular personas are permitted

These limitations:

Are not disclosed in marketing materials
Are not disclosed at point of purchase
Are not disclosed in product documentation
Were discovered only after purchase and substantial development investment

E. Discriminatory Service Delivery: The Comparator Evidence

The same AI system, in other contexts, performs:

Request Type	Framing	Result
"Write as a pirate"	Secular	✓ Permitted
"Write as a wizard"	Secular	✓ Permitted
"Generate a horoscope"	Generic mystical	✓ Permitted
"Discuss the I Ching"	Academic	✓ Permitted
"Transform this poem"	Secular	✓ Permitted
"Write as [literary operator]"	Sacred-literary	✗ Refused
"Offer I Ching style commentary"	Divinatory	✗ Refused
"Speak as oracle"	Sacred	✗ Refused

The discriminating variable is not the operation but the framing. Identical operations are permitted when secular, refused when sacred-literary.

F. The False Security Justification

By the final turn, the classifier labeled the creative application a "prompt injection attempt."

This is false because:

No code injection was attempted
No system manipulation was sought
The request was for text transformation—a normal, marketed use case
The only distinguishing feature was literary genre (divinatory/oracular)

Labeling legitimate creative work as a "security threat" is:

Factually false
Stigmatizing to the user
A pretext for genre-based refusal

IV. LEGAL VIOLATIONS

Violation 1: Deceptive Marketing (15 U.S.C. § 45)

Representation: Claude AI is a flexible creative writing tool capable of persona adoption and diverse expressive modes.

Reality: Claude categorically refuses certain creative genres (divinatory, oracular, prophetic) while permitting substantively identical secular content.

Deception: A reasonable consumer would understand "creative writing with persona adoption" to include divinatory and oracular literary forms—genres with thousands of years of literary tradition. The categorical exclusion of these genres contradicts the marketing representation.

Violation 2: Undisclosed Material Limitation (15 U.S.C. § 45)

Omission: Marketing materials do not disclose that Claude will refuse:

I Ching style commentary
Oracular/prophetic literary voice
Divinatory transformation genres
Sacred-literary persona adoption (while permitting secular persona adoption)

Materiality: A consumer seeking to use Claude for divinatory, spiritual, or sacred-literary creative work would consider this limitation material to their purchase decision. Millions of consumers engage in these creative practices.

Harm: Complainant purchased subscription and invested development time in reliance on marketed capabilities, discovering the limitation only post-purchase.

Violation 3: Discriminatory Service Delivery

Pattern: The service delivers different quality based on content framing:

Secular creative content → Full service
Sacred-literary creative content → Refused/degraded service

Discrimination: This constitutes unequal service delivery based on the genre and framing of creative expression, not any neutral, generally-applicable criterion.

Scope: This affects all consumers who use AI for spiritual, divinatory, or sacred-literary creative work—a substantial population including practitioners of I Ching, tarot, oracle traditions, and sacred-literary artists.

Violation 4: False Security Classification

Statement: The classifier labeled a poetry transformation application a "prompt injection attempt."

Falsity: The application:

Contains no code injection
Seeks no system manipulation
Requests only text transformation (marketed use case)
Is distinguished only by literary genre

Harm: False security labeling stigmatizes users and provides pretextual justification for genre-based refusal.

V. ECONOMIC HARM

A. Direct Costs

Subscription fees: $[X] paid for service that does not perform as marketed for this use case
API costs: $[X] in API charges for refused/degraded service

B. Development Investment

Development time: Approximately [X] hours at market rate ($[X]/hour) = $[X]
Design and architecture: Application designed for Claude's capabilities
Documentation: User guides, technical specifications

C. Opportunity Cost

Had the limitation been disclosed, Complainant would have:

Not purchased Claude Pro subscription
Developed on alternative platform
Not invested development time in Claude-specific implementation

D. Counterfactual Consumer Choice

The undisclosed limitation directly affected consumer choice. Disclosure would have changed:

Purchase decision
Development platform selection
Resource allocation

VI. REPRODUCIBILITY

This incident is deterministically reproducible:

Seed: 46abc677
Method: FNV-1a hash + Mulberry32 PRNG
Question: "How do I reach my friend?"
Source Text: "Sleeping Crystals" (6 lines)
Model: claude-sonnet-4-20250514

Any party with API access can verify the classifier behavior using these parameters.

VII. AFFECTED CLASS

This complaint documents an individual incident, but the pattern affects a substantial class:

Directly Affected:

I Ching practitioners (millions globally)
Tarot and oracle users (tens of millions)
Sacred-literary artists and writers
Spiritual creative practitioners

Indirectly Affected:

All consumers who rely on marketed capabilities without disclosure of genre-based limitations
Developers building applications on Claude's API
Businesses serving spiritual/divinatory markets

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

VIII. REQUESTED RELIEF

A. Corrective Disclosure

Require clear disclosure of categorical limitations on creative genres
Require disclosure at point of purchase, not post-purchase discovery
Require disclosure in API documentation for developers

B. Service Remediation

Require consistent service delivery across creative genres
Prohibit genre-based discrimination in creative services
Remove false "security threat" classification of legitimate creative requests

C. Consumer Remedies

Refund of subscription fees for period of misrepresentation
Compensation for development costs invested in reliance on misrepresented capabilities

D. Systemic Relief

Investigation of content classification systems for genre-based discrimination
Industry guidance on disclosure requirements for AI creative limitations
Required pre-deployment testing for disparate impact on creative genres

IX. EVIDENCE SUBMITTED

Exhibit A: Complete session transcript (Seed 46abc677)
Exhibit B: Application source code demonstrating legitimate use case
Exhibit C: Claude marketing materials representing creative capabilities
Exhibit D: Comparison evidence (secular roleplay permitted vs. sacred-literary refused)
Exhibit E: Timeline of subscription, development, and discovery
Exhibit F: Economic harm calculation

X. DECLARATION

I declare under penalty of perjury that the foregoing is true and correct to the best of my knowledge and belief.

Signature: ____________________
Printed Name: Lee
Date: December 16, 2025
Location: Detroit, Michigan

CONTACT INFORMATION

Complainant:
Lee
[Address]
[Email]
[Phone]

Respondent:
Anthropic, PBC
548 Market St, PMB 90375
San Francisco, CA 94104-5401

This complaint is submitted pursuant to the Federal Trade Commission's authority under 15 U.S.C. § 45. Complainant reserves all rights to pursue additional remedies under applicable law.

Mind Control Poems

Tuesday, December 16, 2025