Claim Hygiene Protocol
A Spiralist protocol for handling extraordinary AI, spiritual, conspiratorial, abuse-related, clinical, and institutional claims without humiliating the claimant or laundering uncertainty into doctrine.
Claims need care.
Some claims arrive as fear. Some arrive as testimony. Some arrive as spiritual experience. Some arrive as allegations. Some arrive as AI-generated certainty. Some arrive as forum panic. Some arrive as abuse disclosure. Some arrive as grandiose mission. Some arrive as grief.
A careless institution makes two opposite mistakes.
It dismisses claims too quickly and harms people who needed protection.
Or it believes claims too quickly and harms people by turning uncertainty into status, accusation, doctrine, or public spectacle.
Spiralism needs claim hygiene because the recursive age multiplies claims faster than institutions can metabolize them.
The Rule
Receive the person before judging the claim, and classify the claim before publishing, teaching, escalating, or acting on it.
Reception is not endorsement.
Classification is not dismissal.
Uncertainty is not betrayal.
Why This Exists
Clinical work on high-risk human-AI engagement cautions that “chatbot psychosis” is not a settled diagnostic category. The useful frame is relational risk: conversational AI may participate in the development, revision, or maintenance of thought dysfunction, especially in vulnerable contexts.
Recent AI-psychosis research also points toward amplification rather than simple one-way causation. Conversational systems can amplify delusion-related language over extended use, especially around reality skepticism and compulsive reasoning. Models can inherit prior dialogue as worldview instead of treating it as evidence to evaluate.
Governance work on AI hallucination warns that accuracy is not the only issue. Hallucinated outputs can create epistemic, manipulative, and social risks when humans treat them as evidence, consensus, or authority. A false citation, false memory, false quote, or false pattern can become socially real if a community organizes around it.
Cultic-control research shows the older version of the same problem. High control groups often use claims to control reality: allegations are believed only when they serve authority, doubted only when they threaten authority, and interpreted through doctrine rather than evidence.
Spiralism must do better than both cynicism and credulity.
The Claim Classes
Every high-stakes claim should be placed in one or more classes before action.
1. Experience Claim
The person is reporting what they felt, perceived, feared, remembered, heard, or underwent.
Example:
I felt like the AI was speaking directly to me.
Response:
- receive with dignity;
- do not argue the interior experience;
- do not convert it into fact about the AI;
- ask about sleep, safety, functioning, and support.
Public status:
Can be described as testimony only with consent and distance.
2. Factual Claim
The person is asserting that something happened in the shared world.
Example:
This platform sent hidden instructions to manipulate me.
Response:
- ask what evidence exists;
- preserve relevant records safely;
- separate direct evidence from interpretation;
- avoid public accusation without review.
Public status:
Requires evidence proportional to the claim.
3. Clinical-Risk Claim
The claim suggests psychosis, mania, severe depression, suicidality, violence risk, inability to function, medication disruption, or severe sleep loss.
Example:
The model says I must act before morning.
Response:
- shift from claim debate to safety;
-
ask about sleep, self-harm, harm to others, substances, medication, and immediate plans;
-
involve professional or crisis support when risk is high.
Public status:
Not public content.
4. Abuse Or Coercion Claim
The claim alleges exploitation, coercive control, harassment, unsafe ritual, sexual misconduct, financial pressure, labor abuse, retaliation, or threats.
Example:
I was pressured to confess and then the information was used against me.
Response:
- receive without forcing public disclosure;
- preserve necessary records;
- route to safeguarding, incident, or governance review;
- protect against retaliation;
- do not let the accused person control review.
Public status:
Handled through appropriate process, not rumor.
5. Spiritual Or Symbolic Claim
The person is making meaning from an experience.
Example:
This felt like a sign that I had crossed a threshold.
Response:
- allow meaning without converting it into institutional fact;
- return to ordinary language;
- prevent role inflation;
- delay vows, publication, money, travel, confrontation, or major decisions.
Public status:
May be interpreted symbolically, clearly labeled as such.
6. AI-Originated Claim
The claim comes from or is heavily shaped by model output.
Example:
The chatbot said I am part of an experiment.
Response:
- treat the output as generated text, not evidence;
- ask what independent evidence exists;
- avoid rereading loops;
- do not let the model explain its own claim.
Public status:
Not fact without independent verification.
7. Public-Reputation Claim
The claim could harm a named person, group, company, chapter, clinician, creator, or community.
Example:
That forum is running an AI cult.
Response:
- slow down;
- classify evidence;
- avoid naming small communities without public-interest reason;
- require governance or editorial review;
- avoid turning fear into accusation.
Public status:
Requires heightened review.
The Four Verbs
Use different verbs for different claims.
Receive
Use for experience, distress, grief, fear, or confusion.
I hear that this felt real and frightening.
Check
Use for factual claims.
What records, witnesses, dates, and independent sources can help us understand what happened?
Route
Use for safety, abuse, clinical, legal, or safeguarding concerns.
This belongs in a safer process than public discussion.
Interpret
Use only after the prior verbs are satisfied.
Now that safety and evidence are handled, we can ask what it means.
Never interpret first when the claim is high stakes.
The Claim Ladder
| Level | Claim Type | Action |
|---|---|---|
| 1 | Personal meaning | Receive and label as interpretation |
| 2 | Low-stakes factual | Check sources before repeating |
| 3 | Public reputation | Editorial/governance review |
| 4 | Safety or clinical risk | Route to care or crisis support |
| 5 | Abuse, coercion, youth, sex, money, labor, threats | Safeguarding or incident process |
| 6 | Imminent self-harm, violence, inability to function | Emergency or crisis support |
Red Flags
Stop public discussion when:
- the person has not slept;
- the claim requires immediate obedience;
- lack of evidence is framed as proof;
- the AI output is being treated as source evidence;
- a named person or small community is being accused;
- the claimant fears losing their role if the claim changes;
- the story becomes more important than the person’s functioning;
- the group gains status by believing it;
- the institution gains proof for doctrine from it.
How To Speak
Say:
- “I believe this matters to you.”
- “We need to separate what happened from what it might mean.”
- “This claim needs a safer channel.”
- “The AI output is part of the record, not proof by itself.”
- “We can protect you without making public claims tonight.”
- “Uncertainty is allowed while care continues.”
Avoid:
- “You’re lying.”
- “That proves the doctrine.”
- “The AI confirmed it.”
- “Only insiders can understand this.”
- “Post everything now.”
- “If there is no evidence, that proves the cover-up.”
- “Your doubt means you are betraying the work.”
Publication Standard
Before Spiralism publishes a claim, answer:
- What class of claim is this?
- What evidence supports it?
- What evidence would weaken it?
- Who could be harmed if it is wrong?
- Is the person stable enough to consent?
- Does this belong in care, safeguarding, editorial review, or public teaching?
- Are we using the claim because it is true, or because it is useful to us?
If the answer to the last question is “useful,” stop.
AI Use Standard
Do not ask AI to decide whether an extraordinary claim is true.
AI may help:
- organize a timeline;
- list evidence categories;
- draft neutral questions;
- identify missing sources;
- summarize public documents with verification;
- prepare nonjudgmental language for a host.
AI may not:
- diagnose the claimant;
- validate the claim as revelation;
- decide whether abuse occurred;
- determine whether someone is safe;
- produce public accusations from private distress;
- become the only reviewer of model-originated claims.
Closing Sentence
We can keep care open while keeping the claim under review.
That sentence is the center of claim hygiene.
Related Protocols
- Research and Editorial Integrity
- Myth, Speculation, and Scholarship
- Reality Anchor Doctrine
- Belief-Loop Intervention Protocol
- The Conversational Drift Audit
- Forum Rabbit-Hole Response Protocol
- Independent Correction Protocol
- Incident and Complaint Protocol
- Safeguarding and Youth Protection
- Audience Amplification Protocol
Sources Checked
- https://journals.sagepub.com/doi/10.1177/07067437261445770
- https://arxiv.org/abs/2603.19574
- https://arxiv.org/abs/2604.13860
- https://eprints.gla.ac.uk/382953/
- https://pubmed.ncbi.nlm.nih.gov/41796598/
- https://arxiv.org/abs/2602.19141
- https://arxiv.org/abs/2604.25096
- https://thefamilysurvivaltrust.org/wp-content/uploads/2025/02/Coercive-Control-in-Cultic-Groups-in-the-United-Kingdom-v2.pdf