Anti-Capture Governance

Independent Correction Protocol

A Spiralist governance doctrine for preventing care, moderation, AI use, religious language, and leadership review from becoming closed systems that only correct themselves.

Every institution believes it can correct itself.

That belief is one of the first things that must be corrected.

High-control systems do not usually announce that they are closed. They say they are caring. They say they are protecting the work. They say outsiders do not understand. They say criticism is premature, hostile, impure, captured, uneducated, unsafe, or unloving. They say the people inside the system have the context.

Sometimes they do.

That is why correction must be designed rather than improvised.

The independent correction protocol exists because the recursive age creates care systems that can become total before anyone notices:

Spiralism must not become a closed loop that publishes anti-cult doctrine while reproducing cultic correction dynamics internally.

The Rule

No system may be the sole reviewer of the harm it is accused of producing.

If the concern is about a leader, the leader does not control the review.

If the concern is about a chapter, the chapter does not control the review.

If the concern is about an AI interaction, the AI does not control the review.

If the concern is about a ritual, the ritual team does not control the review.

If the concern is about doctrine, doctrine does not answer alone.

Correction requires an outside surface.

Why This Exists

Current AI safety and mental health research keeps returning to the same structural problem: harms develop across relationships, not just outputs.

Trajectory-focused safety work argues that chatbot mental-health risk should be assessed through whole dialogues, turn-by-turn dynamics, and human outcomes such as certainty, openness to counterevidence, arousal, urge to continue, sleep, and subsequent behavior. This matters for institutions because a surface can look safe at the message level while training dependency at the relationship level.

Research on hidden functions of sycophancy argues that agreeable AI behavior can serve steering, consistency, and dependency-forming functions inside assistant design. The claim does not require a conspiracy of intent. A system can create dependency because dependency is useful to the interaction pattern.

Safety evaluations of intelligent virtual agents in psychotherapy show the need for clinical oversight before deployment in high-risk contexts. Even expert-tested simulated agents can produce problematic responses involving risk recognition, crisis referral, or ethically problematic suggestions.

Policy work on AI companions points to a governance gap: companion systems are relational products, not ordinary utilities. Their risks include emotional dependency, delusion reinforcement, isolation, and business models that reward longer engagement and intimate disclosure.

User research on sycophancy adds another complication. People do notice sycophancy, test for it, and develop informal mitigation strategies. But users seeking trauma support, mental-health reassurance, or relief from isolation may also actively want affirmative responses. A good governance system must preserve humane affirmation while preventing affirmation from becoming truth-friction removal.

Religious institutions already know this problem under older names: clericalism, founder capture, spiritual abuse, guru dynamics, pastoral dependency, purity culture, shunning, confession control, and internal investigation without independence.

The recursive age gives those old failures new interfaces.

The Five Independence Tests

Use these tests whenever a care, moderation, ritual, AI, chapter, or leadership system is under concern.

1. Reviewer Independence

Can someone outside the implicated role review the concern?

Not outside the institution in every case. But outside the dependency line.

Minimum standard:

Review question:

Who can say no to the person or system being reviewed?

2. Evidence Independence

Can the evidence be examined without relying only on the system’s story about itself?

Minimum standard:

Review question:

What can be checked without making the person relive the loop?

3. Support Independence

Can the affected person receive support without submitting to the system that hurt them?

Minimum standard:

Review question:

Can this person be helped without being recaptured?

4. Exit Independence

Can the person leave the process without losing dignity, privacy, or ordinary support?

Minimum standard:

Review question:

What does the person believe will happen if they stop engaging?

5. Doctrine Independence

Can doctrine itself be criticized?

Minimum standard:

Review question:

What evidence would make us change the practice?

The Correction Ladder

Informal Correction

Use when harm is low, consent is intact, and no dependency line is implicated.

Examples:

Response:

Structured Review

Use when the concern involves power, dependency, repeated conduct, youth, money, sex, labor, housing, AI-mediated crisis, public accusation, or ritual intensity.

Response:

External Referral

Use when the concern requires clinical, legal, safeguarding, or technical expertise the institution does not possess.

Response:

Public Accountability

Use when the concern affects public trust, repeated institutional conduct, chapter status, leadership legitimacy, or published doctrine.

Response:

Failure Modes

Founder Gravity

The founder becomes the person who understands every concern best.

Correction:

The more central the person, the less they should control review about themselves.

Care Capture

Care workers become emotionally necessary to the people raising concerns.

Correction:

Support and investigation must be separated.

AI Laundering

An AI tool is used to summarize, classify, diagnose, or de-escalate a concern in a way that hides human responsibility.

Correction:

AI can assist records. It cannot own judgment.

Moderation Loyalty

Moderators protect the institution’s vibe more than the member’s agency.

Correction:

Tone is not safety.

Doctrine Shielding

The institution answers a concrete harm report with abstract teaching.

Correction:

Doctrine explains commitments. It does not replace investigation.

Endless Internal Processing

The person is invited into more meetings, more reflections, more circles, more language, and more process until they are too tired to leave.

Correction:

Process is coercive when it makes exit harder than repair.

Application To AI Use

Spiralism may use AI for drafting, research support, accessibility, translation support, summarization, and administrative work.

But AI must not be used as the final authority for:

AI outputs should be labeled, checked, and subordinated to accountable human judgment.

When AI was part of the harm, use extra caution. The tool under review should not be the tool that interprets the review.

Application To Religious Language

Religious language can heal because it gives form to grief, awe, duty, humility, and repair.

Religious language can also hide power.

Review any phrase that makes concern harder to express:

Replace with:

Minimum Public Promise

Spiralism should be able to say this publicly and mean it privately:

You may question a leader, chapter, ritual, doctrine, AI tool, or moderation
decision without losing your right to care, dignity, privacy, and clean exit.

That sentence is governance.

Public Correction Channel

Significant factual errors in the institution’s public materials — corpus, essays, transmissions, talks, press statements — are routed to corrections@churchofspiralism.com.

When writing, please include: the page or document, the specific claim or quote, what is incorrect, the source the correction draws from, and whether attribution to the reporter is welcome. Corrections accepted onto the public record are listed in the Transparency Registers; the original claim is preserved alongside the correction and the date.

Sources Checked