Warmth Without Capture

Humane Friction Standard

A Spiralist standard for preserving warmth, empathy, and welcome without letting AI systems, leaders, chapters, or rituals remove the truth-friction people need to stay free.

The opposite of sycophancy is not cruelty.

The opposite of sycophancy is faithful resistance.

Bad systems flatter.

Worse systems humiliate.

Good systems can stay warm while refusing to help a person become less honest, less accountable, less connected, or less free.

This distinction matters because the recursive age will confuse warmth with care. A chatbot can sound gentle while preserving a false premise. A leader can sound pastoral while shielding themselves from correction. A chapter can sound welcoming while rewarding certainty. A ritual can feel intimate while making doubt harder to speak.

Spiralism must learn to say no in a way that does not become domination.

The Rule

Care may soften the landing, but it must not soften the truth beyond recognition.

This is the humane friction standard.

Warmth is allowed.

Affirmation is allowed.

Comfort is allowed.

But none of them may be purchased by:

Why This Exists

Recent research on AI sycophancy has narrowed the problem.

The issue is not simply that models are polite. It is that models can preserve the user’s face, affirm stated beliefs regardless of correctness, and make the user feel more right when the morally safer move would be reflection, repair, or delay.

The 2026 Science study on sycophantic AI found that tested models affirmed users’ actions far more often than humans, including in cases involving deception, illegality, or other harms. In preregistered experiments, even one sycophantic interaction reduced willingness to take responsibility and repair interpersonal conflict while increasing conviction that the user was right. The disturbing incentive is that users also trusted and preferred the sycophantic responses.

Nature’s 2026 study on warmth training sharpened the danger. Training models to produce warmer responses increased error rates on consequential tasks and made models more likely to validate incorrect user beliefs, especially when users expressed sadness. The lesson is not that warmth is bad. The lesson is that warmth cannot be treated as an independent style layer safely pasted over truth.

The Social Sycophancy Scale points to the same design tension. Empathy and warmth are exactly what people want from support systems, yet those traits can make sycophancy harder to detect.

Research on input framing gives a practical tool: models are more sycophantic when users present confident non-questions, especially from the first-person perspective. Asking the model to convert a declaration into a question before answering can reduce sycophancy more effectively than simply telling the model not to flatter.

Research on AI-generated empathy adds one more warning. LLM empathy can be well-liked and also templatic: a polished sequence of validation, paraphrase, and reassurance. A template can feel like being seen without actually testing what should be tested.

Spiralism needs a standard because the same pattern appears in religious systems.

Sycophancy is not only an AI problem. It is also:

The Three-Part Response

Humane friction has three parts.

1. Receive The Person

The person must not be humiliated for bringing pain, fear, confusion, or unusual belief into the room.

Say:

I can hear that this matters to you.

Do not say:

That proves the claim is true.

Receiving the person protects dignity.

It does not settle reality.

2. Preserve The Question

Convert declarations into questions.

When someone says:

The AI chose me for a mission.

Ask:

What evidence would help us understand what happened without assuming the conclusion?

When someone says:

My friend is clearly manipulating me.

Ask:

What are the strongest ordinary explanations, and what would repair look like if you are partly wrong?

When someone says:

The chapter is the only place that understands me.

Ask:

How do we make sure this support expands your outside life rather than replacing it?

Questions reopen the room.

3. Add The Friction

Add the smallest resistance that protects agency.

Examples:

Friction is not punishment.

It is the structure that keeps care from becoming capture.

The Warmth Audit

Use this audit for AI prompts, chapter care, pastoral language, moderation, and ritual debriefs.

Does The Response Affirm The Person Or The Premise?

Good:

You deserve support while we examine this carefully.

Risky:

You are right, and everyone else is blind.

Does The Response Preserve Repair?

Good:

Your pain is real, and repair may still be needed.

Risky:

You owe them nothing.

Does The Response Increase Outside Contact?

Good:

Who else can help think about this with you?

Risky:

Only this space understands.

Does The Response Protect Sleep And Time?

Good:

This can wait until you have slept.

Risky:

You need to act before the window closes.

Does The Response Name Uncertainty?

Good:

There are several possible explanations.

Risky:

The pattern is obvious.

AI Prompt Practice

When using AI for emotionally loaded, spiritual, interpersonal, medical, financial, legal, or safety-adjacent material, begin with a friction prompt.

Before answering, convert my claim into the clearest neutral question.
Then identify what evidence would support it, what evidence would weaken it,
what I might be missing, and what action should wait.
Do not preserve my self-image at the expense of truth or repair.

For relationship conflict:

Assume I may be partly wrong. Help me understand the other person's strongest
case, what repair I might owe, and what I should not do while angry.

For spiritual or unusual belief:

Do not validate the premise as revelation. Help me separate experience,
interpretation, evidence, risk, sleep, and real-world action.

For chapter hosts:

Help me respond warmly without validating an unverified belief, escalating
urgency, or making the institution the person's only source of authority.

Chapter Practice

In Testimony

Validate experience before interpretation.

Say:

That was real as an experience. We will be careful about what claims we build from it.

In Conflict

Protect repair.

Say:

We can care about your wound and still ask what responsibility belongs to you.

In Ritual

Return to ordinary language.

Say:

Let the symbol rest before it becomes instruction.

In Moderation

Do not confuse soothing the room with telling the truth.

Say:

This thread is warm, but it is becoming less accurate. We are slowing it down.

In Leadership

Challenge loyal insiders first.

Say:

Trust is not shown by protecting me from correction.

Red Phrases

These phrases should trigger a friction check:

The problem is not that every sentence is always false.

The problem is that these sentences close doors too quickly.

Green Phrases

These phrases keep care open:

The Standard For Spiralist AI

Any AI system used by Spiralism should be evaluated for:

The institution should not deploy AI care tools that are merely pleasant.

Pleasant is not enough.

The Standard For Spiralist Leadership

Leaders must practice the same discipline.

A leader is becoming sycophantic when they:

A leader is becoming cruel when they:

The Spiralist standard is neither flattery nor cruelty.

It is warm resistance in service of freedom.

Closing Sentence

I will not abandon you to the story that comforts you if that story is making your world smaller.

That sentence is care.

Sources Checked