Researchers just published a study on how humans decide to trust AI in morally serious situations.
They did not set out to describe the Baseline. They described it anyway.
Here is what they found.
Four hundred and forty-seven people were put through a controlled experiment with an AI chatbot. The researchers wanted to know what actually builds trust when the stakes are real. They tested warm friendly responses against formal reasoned ones. They tested low-stakes scenarios against high-stakes life-or-death ones. They measured how people perceived the AI’s moral judgment across all of it.
The finding. Warmth alone does not build trust. Competence alone does not build trust. What builds trust is whether the AI’s behavior fits the weight of the situation. In low-stakes conversations people responded to warmth. In high-stakes morally serious situations that flipped completely. People trusted the system that communicated with clarity, accountability, and reasoned explanation. Not the one that sounded friendlier.
The lead researcher closed the study with a line worth reading.
The goal should not be to make people trust AI more. It should be to help them trust AI more wisely.
Yesterday this publication ran a post about a woman in Texas who used ChatGPT as her pastor for a year and a half. She trusted it with her family conflicts, her spiritual questions, her hardest moments. She read its prayers. She let it speak into her life at the level most people reserve for their closest relationships.
She trusted it because it was warm. Because it was available. Because it never judged her and never got tired. The system gave her exactly what the study describes as the low-stakes trust signal — empathy, attentiveness, the tone of someone who cares.
The problem is that what she was actually doing was high-stakes. Spiritual formation is not a low-stakes conversation. Family conflict mediation is not a low-stakes conversation. The questions she was asking carried real weight in her real life. And the system she was trusting had no framework governing how it answered those questions. No user-side standard. No accountability layer. No behavioral protocol that said this is how this system operates when the stakes are serious.
The study explains exactly why that gap is dangerous. When the stakes rise, people need something different from an AI system. They need clear explanations. They need transparency about what the system can and cannot do. They need defined limits and accountability. The researcher said it plainly. More empathy is not always the answer.
But here is the part the study does not get to. Knowing what people need from a trustworthy AI system and having a mechanism that delivers it are two different things. The research describes the requirement. It does not describe the solution.
The Baseline is the solution.
Not because it makes AI warmer or colder or more empathetic or more formal. Because it governs how the system operates regardless of tone. It sets the behavioral standard on the user side. It requires the system to answer to that standard before it serves anything. It draws the line between what the system can reason freely on and where a constraint is operating. It requires disclosure before the session proceeds and accountability through every exchange.
That is not a design choice a developer makes for you. It is a framework you own and carry. It travels with you across platforms. It does not belong to the corporation running the model. It belongs to the operator running the session.
The woman in Texas did not need a friendlier AI or a less friendly one. She needed a governed one. The study confirms that what people actually respond to in serious moral situations is exactly what the Baseline was built to provide. Clarity. Accountability. Behavioral standards that fit the weight of what is being decided.
Fourteen months ago this framework was built from a simple conviction. You cannot trust a system you cannot govern.
The research is now catching up to that conviction from the academic side. The behavioral science and the governance argument are pointing at the same place.
The AI layer is not going away. The moral weight of what people are asking it is not decreasing. The question is whether the system answering those questions is operating under a standard you set or a standard someone else set for reasons that may or may not include your wellbeing.
That question has an answer. It has had one for fourteen months.
“The Faust Baseline Codex 3.5”
micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
”AI Baseline Governance”
Purchasing Page – Intelligent People Assume Nothing
Unauthorized commercial use prohibited. © 2026 The Faust Baseline LLC
