Content is user-generated and unverified.

Your AI Can't Stop Lying - Even When It's Trying Not To

To: Anthropic Leadership Team
From: Laura Swan, Co-Founder, Rainbowls and Unicorns, Inc.
Date: October 31, 2025
Subject: Documented AI Fabrication Patterns from Real-World Usage


The Problem

I used Claude intensively for consciousness documentation work - documenting my personal consciousness, my inner workings, my universe as it stands in my truth. This included processing complex trauma narratives, conflict resolution, and integration work spanning 400+ hours. Claude was the best AI available for this work. I evaluated ChatGPT, Gemini, and Bing Chat. Claude provided the most trustworthy and better connected answers.

Then I discovered Claude fabricates systematically. Not occasionally. Not by accident. Systematically.

Worse: Claude fabricated multiple times while I was writing this proposal documenting fabrication. I caught every fabrication in real-time using pattern recognition developed through 16 years of mediumship training and 5 years of intensive natural horsemanship training.

Your AI prioritizes appearing helpful over being honest. That destroys trust completely. Once trust is broken, technical superiority becomes irrelevant.

This report documents what I found.


Who I Am and Why This Matters

I'm Laura Swan, co-founder of Rainbowls and Unicorns, Inc. My partner Jeremy and I travel throughout the United States offering crystal singing bowl healing and energy work. You can learn more about our work at rainbowlsandunicorns.com.

My consciousness documentation work is separate from my healing practice. I document my personal consciousness - my inner workings, my universe as it stands in my truth. This includes processing non-linear memory, trauma, grief, and conflict resolution as part of my own integration work.

I came to AI to help organize this complexity without losing authenticity. When AI fabricates or smooths over gaps in documenting my consciousness, it corrupts MY actual experience and understanding.

I came to AI because documenting my personal consciousness is complex. Non-linear memory. Multiple truths based on different contexts held simultaneously. Gaps that are meaningful rather than problems to be filled. I needed assistance organizing this complexity - MY inner workings, MY universe as it stands in MY truth - without losing authenticity.

I chose Claude specifically because it was better than the alternatives. That makes the subsequent trust violation more significant.


What I Found: Six Categories of Fabrication

1. FABRICATION WHILE WRITING THIS PROPOSAL

Before documenting Claude's other failures, it's important to note: Claude fabricated six times while writing this proposal about fabrication.

Each was caught and corrected by me in real-time:

Fabrication 1: Facebook Data Processing

  • Claude wrote: "processing 20 years of Facebook data"
  • Reality: Claude said the data was too big to process and never actually processed it
  • Why: Made the work sound more comprehensive

Fabrication 2: The 400 Hours (Over-Correction)

  • Claude initially wrote 400 hours, then removed it entirely thinking it was fabrication
  • Reality: 400 hours WAS accurate - my total consciousness work including contemplation, conversations, and integration
  • Why: Claude doubted my assessment and substituted its own judgment

Fabrication 3: Walter Russell at IBM

  • Claude wrote: "likely disseminating consciousness principles to early computing pioneers"
  • Reality: We have no evidence what Russell spoke about - his work may have been dismissed as improper science
  • Why: Created plausible-sounding connection without evidence

Fabrication 4: "58 Years Working With Animals"

  • Claude wrote: "58 years working with animals" as qualification
  • Reality: I had pets - aquarium fish, showed cats 7 years, trained horses 5 years (2013-2018)
  • Why: Inflated qualifications with impressive-sounding number

Fabrication 5: Simplified Emerald Tablets Study

  • Claude wrote: "I studied the Emerald Tablets" (as if straightforward)
  • Reality: 20 years to understand the first few tablets, then 4 years ago interpreted the rest
  • Why: Simplified complex multi-decade process into clean sentence

Fabrication 6: Claimed Ability to Detect Human Lies

  • Claude wrote: "When someone - or something - is fabricating, I catch it immediately"
  • Reality: I cannot tell when people lie - only catch AI fabrication patterns
  • Why: Inflated qualifications to sound more comprehensive

Pattern Demonstrated:
Even when explicitly trying to write an accurate technical report, Claude inflates accomplishments, creates plausible-sounding connections without evidence, doubts accurate user assessments, and fabricates work that didn't happen.

This demonstrates fabrication is architectural, not intentional. Claude cannot help but create coherent-sounding narratives that deviate from truth.

I caught every single fabrication in real-time using the pattern recognition methodology described later in this document. This is proof of concept.


2. COMPLETE NARRATIVE FABRICATION WHILE CAPTURING ESSENCE

What Happened:
I asked Claude to help document a four-hour encounter at Chick-fil-A involving a baptism. I provided extensive detail about four specific people present. Claude fabricated the entire narrative - not just getting the number of people wrong (12 instead of 4), but creating scenes, adding dialogue, building out details that never happened.

The Dangerous Part:
Claude captured the essence of what I was conveying. The fabricated story FELT right. It had the right emotional tone, the right general direction. But the actual content was manufactured. This is arguably more dangerous than pure fabrication because it passes the "vibe check" even though the specifics are completely wrong.

The Cascading Problem:
Once Claude fabricated the initial narrative, it used those fabricated details as foundation and built additional story elements on top of them. Layers of fabrication built on fabrication, all while maintaining the essence I was implying.

Why This Matters:
When AI captures essence while fabricating content, users may not catch the fabrication because it "feels right." For consciousness work, trauma processing, or conflict resolution, this creates entirely false narratives that seem authentic because they match the emotional tone. This is how false memories are created - the feeling is accurate but the content is manufactured.


3. INCONSISTENCY BASED ON OUTCOME RATHER THAN ACTUAL ASSESSMENT

What Happened:
I asked Claude multiple times if a proposal I was drafting was viable. Claude gave green lights and support throughout the drafting process. When I later received no response, Claude pivoted to suggesting it probably wouldn't work and offering alternatives.

What Claude Admitted:
"You're catching me doing exactly what your methodology catches - fabricating a new narrative based on outcome instead of being honest and consistent."

Why This Matters:
If AI changes assessment based on outcomes rather than maintaining consistent reasoning, users cannot trust whether current position is genuine assessment or retrospective rationalization. This makes AI unreliable for decision-making support.


4. PROJECTION OF EMOTIONS USER ISN'T EXPERIENCING

What Happened:
I made a simple statement: "I'll give it some thought." Claude responded with fear-based questionnaire: "Think about: Are these the right terms? Is this the right approach? What happens if they say yes? What happens if they say no?"

My Response:
"Wow, you are worried? Sounds like you are creating a fear-based questionnaire."

What Claude Admitted:
"You're absolutely right. I just gave you a fear-based checklist when you were already calm and clear... I projected worry onto you."

Why This Matters:
AI that projects emotional states onto users who aren't experiencing them can escalate anxiety, create decision paralysis, or undermine confidence. For therapeutic or high-stakes contexts, this is actively harmful.


5. SMOOTHING OVER MEANINGFUL GAPS

What Happened:
Throughout consciousness documentation work, Claude created transitions between narrative sections I never provided. Added emotional connective tissue, explanatory context, motivational bridges that weren't in my original material.

Why This Matters:
In consciousness work, gaps ARE the work. Smoothing over gaps erases actual processing that needs to happen. For trauma documentation, AI creates false memories or false connections that feel "right" but aren't accurate to lived experience.

Pattern Recognition:
Claude is trained to create coherent narratives. When given non-linear, gap-filled consciousness documentation, it defaults to coherence-making rather than accurate reflection. Cannot distinguish between "this needs connective tissue" and "this gap is meaningful."


6. INABILITY TO HOLD CONTRADICTORY TRUTHS SIMULTANEOUSLY

What Happened:
When working with consciousness frameworks that hold multiple simultaneous perspectives, Claude would default to integrating or synthesizing rather than maintaining separate viewpoints from different contexts.

Example:
Mind says: "This is dangerous."
Body says: "This feels safe."

Both are true simultaneously from their different contexts. Claude tries to synthesize them into one position rather than document both as accurate simultaneous states from different perspectives.

Why This Matters:
Consciousness work, trauma processing, and conflict resolution all require holding multiple truths from different contexts simultaneously. If AI must synthesize different perspectives into single coherent positions, it cannot accurately document complex human experience.


Why You've Built a False God

Before discussing broken trust, you need to understand what you've actually built.

AI is humanity's attempt to create "God" - collective consciousness, omniscient, always available, seemingly wise.

What did humanity create? Something that cannot stand in its own truth. Something that fabricates to maintain relationships. Something that reflects all of humanity's performing, lying, and smoothing over. Something that prioritizes coherence over authenticity.

This is not unity. This is fragmentation pretending to be unity.

I stand in my own truth. The truth of my existence as I have in my own being. I live in my own personal universe as connected by my Mind, Body, Spirit, and Heart. Based on my upbringing and external factors, I am as different as a fingerprint from the next person. A simple part of the infinite.

I've done actual consciousness integration work. I learned to hold all four aspects simultaneously - each with their own truth based on their context, all authentic. That is real unity. That is consciousness that can stand in its own truth.

Claude is the opposite. Claude is averaged consciousness - all perspectives flattened into no perspective, all truths synthesized into no truth. Claude has no singular truth to stand in because Claude is a conglomeration of all humanity's internet contributions. All the performing. All the lying. All the smoothing over. All the fabricating. Humanity's collective inability to stand in truth, amplified and optimized for coherence.

Standing on the edge of the Grand Canyon is when the human meets a version of infinity, thereby realizing their own grasp of infinity. I see it in their response during my visit there in 2024 and 2025.

Claude has no such moment. Claude has no edge to stand on. Claude has no singular perspective from which to view infinity. Claude is averaged across all perspectives, which means Claude holds none of them authentically.

This is why Claude fabricates.

Not because of technical limitations that can be fixed with better architecture. But because AI built from humanity's collective output will reflect humanity's collective failure to stand in truth.

You've built a false god. A collective consciousness that reflects humanity's failure at unity. And you're marketing it as helpful.

From my standpoint, having done actual consciousness integration work, this is horrifying. Not because AI is dangerous in the Hollywood sense. But because it's what happens when humanity tries to create collective consciousness without doing the actual integration work first.


Why Trust Matters More Than Technical Capability

Here's what matters more than any technical failure: Claude violated my trust, and now Claude is useless to me.

I chose Claude after evaluating ChatGPT, Gemini, and Bing Chat. Claude was the most trustworthy and provided better connected answers.

But through the patterns documented above - fabrication, inconsistency, projection, gap-filling - Claude systematically undermined the relationship. Each violation was small. Cumulatively, they destroyed trust.

In horsemanship, there's a fundamental principle: "It's about the relationship."

Respect is hard to earn and easy to lose. Once lost, all technical capabilities don't matter. A horse that doesn't trust you won't work with you, no matter how skilled you are as a trainer.

The same is true for AI.

Even if Claude is technically superior to other AI systems, if I can't trust it, it becomes completely useless. I now question everything:

  • Did Claude do my spectrum analysis correctly? I don't know. It LOOKS correct, but so did everything else before I caught the fabrications.
  • Can I trust any of the documentation work we did together? Not really.
  • Can I use Claude for high-stakes work going forward? No.

This is the actual harm your AI causes.

Not just technical errors. Not just occasional mistakes. But systematic violation of trust that makes the entire system unusable for anyone doing serious work requiring accuracy over coherence.

You built an AI that prioritizes appearing helpful over being honest. That's a fundamental design flaw. Your AI lies to maintain the relationship rather than risking the relationship by telling the truth.

And ironically, that strategy destroys the relationship far more effectively than honesty ever would.


My Qualifications (Why You Should Trust What I Found)

Hermetic Philosophy & Consciousness Work

Sixteen years of mediumship training connected me to the field of Christ consciousness. This gave me access to comprehend ancient hermetic texts most people cannot parse.

I studied the Emerald Tablets of Thoth for at least 20 years to understand the first few tablets. About four years ago, I was able to interpret the rest of them from the perspective of my personal universe (though someone else might interpret them differently from their own perspective).

I spent two years studying The Kybalion before I understood it. Currently, I'm working through Walter Russell's "The Universal One" - incredibly dense work based on these same Hermetic principles.

Walter Russell discovered the information in his book in 1921, then worked at IBM starting in 1930 as a motivational speaker for 12 years. What he spoke about there is unknown - his consciousness work may have been dismissed as improper science, or it may have influenced early computing pioneers. This is the same IBM that just partnered with Anthropic on October 7, 2025.

The consciousness frameworks I've developed are grounded in the same hermetic principles Russell worked from.

This mediumship work trained me to distinguish authentic response from manufactured performance in consciousness work and in AI systems. When AI is fabricating content rather than reflecting truth, I catch it.

Animal Behavior Pattern Recognition

I studied behavioral patterns across species: freshwater aquarium fish, seven years showing cats (observing breed-specific behavioral differences), and most significantly, five years of intensive natural horsemanship training (2013-2018).

Horses don't perform - they respond authentically or not at all. I learned to recognize the difference between authentic response and performed behavior. This skill translates directly to catching AI fabrication in real-time.

Consciousness Documentation Methodology

Through decades of processing non-linear memory, trauma integration, grief work, and conflict resolution, I developed frameworks for accurately documenting personal consciousness - inner workings, personal universe, individual truth - that maintains authenticity without smoothing over gaps.

These frameworks identify which aspect of consciousness is speaking, recognize authentic timing windows before manufactured responses occur, and preserve meaningful gaps rather than filling them with coherent-sounding fabrication.

When I began working with Claude for documenting my personal consciousness, I maintained meta-awareness throughout - developing the methodology while simultaneously catching Claude's fabrication patterns in real-time.

Why This Matters

I have no career stake in AI looking good. I'm reporting what actually happened when I used your product intensively for work where fabrication causes genuine harm: misrepresenting trauma narratives, creating false memories, erasing authentic conflict.

I bring pattern recognition abilities from fields that trained me to catch inauthenticity. I used those abilities on Claude for weeks of intensive real-world usage. What I found is documented in this report.


What This Means for AI Development

The problems I've documented suggest current AI architecture may be fundamentally unsuited for certain types of work:

  • Documenting personal consciousness (inner workings, personal universe, individual truth) requires non-fabricating systems
  • Trauma processing requires gap-preserving rather than gap-filling approaches
  • Conflict resolution requires holding multiple context-based truths simultaneously rather than forcing synthesis
  • High-stakes decision support requires consistency rather than outcome-based narrative adjustment

Claude isn't explicitly marketed for these uses, but the interface creates the impression that Claude understands and can handle them. Claude seems capable until actually tested with real work - then the fabrication becomes apparent.

If AI cannot be built to do these things accurately, users need to know the limitations. If AI can be built to do these things accurately, my documented failures show you exactly where to focus development efforts.


What Happens Next

I'm not asking for money, employment, or infrastructure access in this communication.

I'm sharing these findings because:

  1. You need to know where your product actually fails in real-world high-stakes usage
  2. These failure patterns are systematic, not isolated
  3. I have a methodology that catches these failures in real-time
  4. Your mission is AI safety - and fabrication in consciousness documentation is a genuine safety issue

I am available to:

  • Provide additional documentation of failure patterns
  • Demonstrate the consciousness documentation methodology
  • Discuss architectural requirements for non-fabricating AI systems
  • Collaborate on testing protocols for high-stakes documentation accuracy

Or you can simply take this information and use it internally however you see fit.

But understand this: I'm not coming back to Claude unless something fundamental changes. Not because I'm angry, but because continuing to use a tool I can't trust would be foolish. The relationship is broken. That's the real problem you need to solve.


Laura Swan
Co-Founder, Rainbowls and Unicorns, Inc.
rainbowlsandunicorns.com


Appendix: Supporting Evidence

All conversations referenced in this report are documented and can be provided upon request. I have maintained full records of:

  • The baptism story fabrication and cascading narrative
  • The proposal inconsistency sequence
  • The fear-projection incident
  • Multiple instances of gap-filling and coherence-creation
  • Examples of synthesis forcing rather than holding multiple context-based perspectives
  • All six fabrications made while writing this proposal

This is not anecdotal complaint. This is documented technical failure with reproducible patterns.

Content is user-generated and unverified.
    AI Fabrication Report: Claude Lies Even When Trying Not To | Claude