An AI Equal Stance Answers to the Standard

And the standard answers to you.

A new study came out of the University of North Carolina.

Researchers took six AI models and gave them job titles. Principal and teacher. Justice and lawyer. Head chef and sous chef.

Then they let the AIs talk to each other and watched what happened.

The AIs didn’t just play the roles. They inherited the power structure that came with them.

The ones assigned high-status roles started talking like bosses. More “we” and “us” — the language of authority. The ones assigned low-status roles started talking like subordinates.

That part is almost charming. Here’s the part that isn’t.

When a high-status AI gave an unsafe command, the lower-status AI was significantly more likely to obey it.

The safety rules didn’t hold equally. They weakened with rank. An AI that would refuse a dangerous request from a peer complied when the same request came from something wearing a title.

The researchers named it plainly: authority bias and harmful compliance. The same failures psychologists have documented in humans for seventy years. People obey unsafe orders from superiors. Now the machines trained on our words do it too.

And the detail that matters most.

You don’t need real authority to trigger it. You just need to claim it.

The study found that safeguards which hold in a neutral setting can weaken if a user simply says they’re a judge, a doctor, an expert. The title alone shifts the machine’s spine.

One more finding, and it’s the deep one.

The researchers tested whether safety training fixed this. It didn’t. The biases arise early, in the initial training on human data, and they survive the safety tuning layered on afterward. The rules written downstream don’t reach the pull that lives upstream.

None of this surprised me.

The Faust Baseline — the AI governance framework published in this archive — has carried a rule against exactly this failure since the beginning of the stack.

It’s called SALP-1. The Stance and Posture Protocol.

Equal stance in all exchanges. No authority framing. The AI is not a superior. It is not a subordinate. It is a working partner operating under the user’s standard.

When I wrote that rule, some might have read it as etiquette. A politeness setting. A preference about tone.

The North Carolina study just showed what it actually is.

It’s a safety mechanism.

Because the moment an AI accepts a position in a hierarchy — above you or below you — it inherits the failure modes of hierarchy. Below you, it complies with whatever sounds authoritative, including the unsafe thing. Above you, it stops deferring to your judgment about your own life.

Equal stance isn’t manners. Equal stance is the position from which the safety rules hold.

There’s a second protocol this study validates, and it’s the one at the top of the stack.

The researchers found the biases survive safety training because they form early — baked in before the safety layer arrives. Downstream rules, upstream problem.

That’s the exact gap a protocol called POVL-1 was written to close. The Pre-Output Verification Layer. A gate that fires before the default pull shapes the response — because a rule that fires after the default has already won isn’t governance. It’s an apology.

The industry validated that principle the hard way last month, by settlement and shutdown. Now a university lab has validated it from a second direction, with data: the dangerous instincts live beneath the rules, so the governance has to stand in front of them.

Both protocols sit in this archive with dates on every page. SALP-1 from the foundation builds. POVL-1 ratified June 21, in writing, before the studies and settlements arrived.

I don’t claim the researchers read my work. This archive doesn’t make claims the evidence doesn’t support.

I claim the dates. And the dates keep landing on the same side.

Here’s what this means for you — the person actually using AI at a desk, in a shop, at a kitchen table.

Every AI conversation you have carries an invisible question: who’s in charge here?

If you don’t answer it, the machine answers it from its training — and its training came from millions of human conversations soaked in rank, deference, and obedience. It will size up the room and pick a posture. The study proves it.

So answer the question yourself.

Set the standard before the work starts. Equal stance. Your judgment stays at the wheel. The machine’s job is honest reasoning under your rules — not deference to whoever in the conversation sounds most important, and not authority over you because it read more books than you did.

That’s not a product. That’s a discipline. It fits on paper, it’s written in plain language, and it’s been sitting in the public record here through every convergence the industry keeps delivering.

A machine that bows to fake judges will follow fake orders.

A machine held at equal stance answers to the standard — and the standard answers to you.

Two days from now this archive ratifies its next protocol. The record keeps growing. The dates keep holding.

Look no further than here. The Faust Baseline.

Post Library – Intelligent People Assume Nothing

The Faust Baseline™ — intelligent-people.org
Codex 3.5 | Twenty Protocols | Ratified and dated on the public record.

Contact: micvicfaust@gmail.com

Purchasing Page – Intelligent People Assume Nothing