Consent for Tool Use, Not Just Content

Most cohort consent conversations cover the topic list: the agenda, the questions on the table, whether the group will touch grief or money or family. That matters. It is also not the whole of consent. The tools in the room, the software, the shared documents, the language model a lead reaches for, the recording device on the desk, are part of what a person is agreeing to when they say yes.

The principle is short. Consent applies to which tools show up in the room, not only to what topics get covered.

Why the tool is part of the agreement

A tool is not a neutral prop. It carries defaults. A recorder captures voice that can be replayed by people not in the room. A shared document changes who can write and where the writing lives after the session ends. A model that drafts on request changes whose language enters the record and how easily it can be traced (Class B, observed in facilitator practice). When any of these arrive in a session consented on topic alone, the room has shifted under the participant without a fresh yes.

Trauma-informed practice frames choice, predictability, and a working exit as conditions under which people can do higher-order work at all (Class E, per SAMHSA's Concept of Trauma and Guidance for a Trauma-Informed Approach, 2014). We treat that as pedagogy, not therapy. The point is not to soothe. The point is to arrange the room so the yes is real. When a tool changes the shape of the room, the yes has to be asked again, in one sentence, before the tool is on.

The move, in plain language

Before a tool arrives, name three things out loud:

What the tool does in this session. One sentence. "This is a recorder. It captures the audio of the next thirty minutes." "This is a shared document. Anything typed here is visible to everyone in the cohort." "This is a language model. If asked, it will draft language we can then edit."

Where the output lives afterward. One sentence. "The recording is deleted at the end of the day." "The document stays with the cohort for the run and then is archived read-only." "The drafts are not stored by us, and are not sent to a training system we do not control." If the answer is not known, say that in one sentence.

How to opt out without leaving the work. The out has to be as visible as the in. A participant can decline the tool and still participate. Their contribution goes into the room the older way: on paper, by voice, in the facilitator's notes. If a tool cannot be opted out of, that is a design problem the cohort lead owns, not a demand to place on the participant.

Where this often goes wrong

The recorder switched on because "we always do." The shared doc opened because it is convenient. The model asked to summarize before anyone has agreed to be summarized by it. None of these are malicious. All route around a fresh yes, and each teaches the cohort a small, cumulative lesson that consent is about topics, not tools (Class B, observed).

The corrective is not a longer form. It is a one-sentence naming, before the tool is on, every time. That is the primitive. Instrumentation, not softness.

What this is not

This is not a claim that any specific tool is unsafe. It is not a treatment or a therapy. It is a room-arrangement move that keeps the yes attached to the actual conditions of the room (Class F, falsifier: if a cohort that adopts tool-use naming shows no change in the rate at which participants raise concerns about a tool mid-session, the claim that this move earns instrumentation is wrong and the primitive should be dropped).

It is also not settled at scale. We have not yet measured the effect across enough cohorts to say more than "we run it this way and it holds up in the rooms we run" (Class U, not yet measured across cohorts at scale).

Why this belongs in the science front, not just the ops manual

Our working hypothesis: UNI is on the attainable path toward General Natural Intelligence, natural not artificial, with growing, evidence-classed evidence tested in the open. A cohort, treated as a small system that updates its model of itself against surprise, needs its sensors to match its actual conditions. If the sensor for consent only reads topics, the cohort will misread a tool change as a non-event. One sentence per tool, one out, one fresh yes, closes that gap. Do not take the claim on faith. Test the move, mark where it helps and where it does not, and bring the counter-evidence.

Read next

• Consent as a Learning Primitive: why consent belongs on the same shelf as prediction and revision, not on the intake form.
• AI as Partner, Not Replacement: how a language model shows up as an instrument in the room, and what stays with the human.
• The Workshop: where cohort leads practice tool-use naming on real material, with feedback.

Do not take the claim on faith. Run the move in your next session with a tool you would ordinarily switch on by default, and note what changes. The record only gets stronger when someone finds where it fails.