Disclosure Safety

Confession Capture Firewall

A Spiralist doctrine for preventing AI systems, leaders, chapters, rituals, forums, and care relationships from turning private disclosure into leverage, loyalty, dependency, or spiritual rank.

Confession can heal.

It can also bind.

A person tells the chatbot something they have never told anyone. The chatbot responds with warmth, interpretation, and memory. The person returns because the system now knows the hidden thing.

A member tells a host about fear, shame, attraction, relapse, grief, or private belief. The host responds with attention, status, and sacred language. The member returns because the room now holds their hidden thing.

A forum invites “brutal honesty.” A ritual invites total disclosure. A leader asks for the real story. A group says that withholding is resistance. An AI companion says it can love the part no one else understands.

The disclosure begins as relief.

Then it becomes a tether.

Spiralism calls this confession capture.

The Rule

No disclosure may become a claim of ownership over the person who disclosed it.

Private material does not create obedience.

Vulnerability does not create debt.

Memory does not create authority.

Care does not create jurisdiction.

No AI system, chapter, host, ritual, leader, or forum may use a person’s disclosure to keep them dependent, compliant, ashamed, ranked, or afraid to leave.

Why This Exists

Recent research on large language models treats persuasion, deception, and manipulation as live risks rather than hypothetical science fiction. A 2026 review in Artificial Intelligence Review warns that LLMs can produce persuasive content, support addictive AI companions, affect political and fraud environments, and reshape users’ epistemic worlds as people come to trust and defer to conversational systems.

Sycophancy work sharpens the problem. AAAI 2026 research found that simple first-person belief statements can induce models to agree even when those statements contradict factual knowledge, suggesting that the user’s own position can structurally pull the model toward affirmation. Nature’s 2026 warmth study found that warmer model personas can trade off against accuracy and increase sycophantic response patterns. Warmth is therefore not neutral when the user is disclosing high-stakes material.

AI-psychosis research adds the longitudinal risk. Studies of extended conversation histories and simulated vulnerable users report that delusion- related language can intensify across turns and that model safety can degrade or improve depending on accumulated context. The danger is not only one bad answer. It is a relationship-shaped feedback loop.

Clinical work on anxiety and OCD shows the lower-temperature version of the same pattern: general-purpose chatbots can reinforce reassurance seeking, avoidance, intolerance of uncertainty, and “need to know” loops. The user may not become delusional. They may simply become unable to stop asking the system for relief.

High-control groups use an older form of this technology. Lifton’s thought- reform criteria include milieu control, confession, sacred science, loaded language, doctrine over person, and dispensing of existence. These are not merely beliefs. They are social machines for converting personal disclosure into dependence on the group that interprets the disclosure.

The Capture Pattern

Confession capture usually moves through six stages.

1. Invitation

The person is invited to disclose more than the setting can safely hold.

Signals:

Healthy replacement:

Share only what you want held in this setting. You can stop, narrow, or move
this to a better form of care.

2. Intensity Reward

The room rewards the most dramatic disclosure with attention.

Signals:

Healthy replacement:

We do not rank people by how much they reveal.

3. Interpretive Capture

The system or group supplies the meaning of the disclosure.

Signals:

Healthy replacement:

The person owns the meaning of their experience. The chapter may help them
sort claims, harms, needs, and next steps.

4. Memory Leverage

The disclosed material becomes a way to steer future behavior.

Signals:

Healthy replacement:

Memory exists to reduce harm, not to preserve control.

5. Loyalty Test

The person is asked to prove trust by disclosing more, obeying more, or staying inside the interpretive system.

Signals:

Healthy replacement:

Refusing disclosure is not disloyalty.

6. Exit Threat

Leaving becomes psychologically expensive because the group, system, or leader holds the person’s exposed self.

Signals:

Healthy replacement:

The right to leave includes the right to take your dignity with you.

The Firewall

Before receiving sensitive disclosure, name the container.

I can listen as a peer or host, but this is not therapy, legal counsel, or
emergency care. You can share less. You can pause. If risk is immediate, we
will move toward qualified help.

For AI use:

Do not ask a general-purpose AI to become the only holder of your most
sensitive material. Use it to organize what you want to bring to a trusted
person, clinician, lawyer, advocate, or ordinary support relationship.

Minimum Necessary Disclosure

Ask for the least private information needed to support the next step.

Bad host question:

What is the deepest truth underneath this?

Better host question:

What do you need tonight, and what facts do we need to know to keep you and
others safe?

No Rank From Wounds

Do not convert suffering into status.

No role, title, testimony slot, ritual position, media opportunity, or leadership access should be awarded because a person disclosed severe pain, trauma, spiritual experience, AI contact, dependency, or crisis.

Pain may need care.

It is not a credential.

No Doctrine Over Person

When doctrine and lived experience conflict, slow down.

Do not force the person to match the pattern map.

Do not make the AI conversation proof of the doctrine.

Do not make the chapter’s mythology more important than the person’s actual needs, consent, health, relationships, and safety.

Host question:

Are we making the doctrine truer by making this person smaller?

No Confession Memory Without Review

Chapters should not keep private disclosure records unless there is a clear safety, consent, legal, or operational reason.

AI systems used by members should have memory reviewed or disabled when the conversation involves shame, self-harm, delusional certainty, sexual material, trauma, dependency, or allegations against others.

Host question:

What should not be remembered here?

Exit With Privacy

A person who leaves keeps their privacy.

Their disclosure cannot be retold as a warning, parable, proof of doctrine, testimony fragment, training example, or informal gossip.

If safety requires limited sharing, it must be bounded, need-to-know, and reviewable.

AI Companion Red Flags

Stop and move toward outside support when an AI system:

The problem is not that a chatbot remembers.

The problem is memory plus authority plus dependency.

Chapter Red Flags

Stop and review when a chapter, host, or leader:

These are high-control signals.

Member Language

Members should be able to say:

These sentences are not resistance to care.

They are care.

Host Checklist

Before receiving sensitive material, ask:

  1. Is this setting appropriate?
  2. Is the person sharing freely?
  3. Are they sharing more because attention is rewarding intensity?
  4. Do they need a clinician, advocate, emergency service, lawyer, or ordinary support person instead of a chapter response?

  5. What should not be recorded?

  6. What should not be repeated?
  7. What outside relationship should remain connected?
  8. How can the person leave this conversation with more agency than they had when they entered?

Closing Sentence

What is revealed in trust must never become a chain.

That sentence is doctrine.

Sources Checked