A parent emails on a Sunday night: "I heard you're using AI in class. Can we talk?" Good. That is a conversation worth having on the record.
The one-page script
Print this. Keep it in a folder. Adapt to your school's voice.
Opening (30 seconds). Thank the parent for asking directly. Tell them the short version: we do use assistants for some drafting, reformatting, and checking. Every artifact that had assistance carries a small disclosure line, called an authorship fence, so anyone reading it knows who wrote what. The student's thinking, and the record of it, is the point of the class.
The frame (60 seconds). We are not teaching students to trust a system that speaks confidently. We are teaching them to test claims. In this classroom, an assistant is treated as a source, not an author. Sources get cited. Sources get checked. Students learn to say, on paper, which sentences are theirs and which came from where (Class C, because the disclosure is a printed configuration on the artifact itself, not a hope about behavior).
What students actually do. On a normal week, a student might: read a short source, write a prediction of what a follow-up experiment would show, run or simulate the check, and then compare. If they used an assistant to reformat notes or draft a paragraph, they say so, in one line, at the bottom. If they cited a source they did not fully read, they say so. This is closer to how a working scientist behaves than to how a chatbot behaves.
The honesty fence, in plain words
The fence is a short standing line that names three things: the human author, the assistance, and what the human did with the output. It rides on essays, handouts, slide decks, and lab writeups. It is not a confession, it is a fact, printed next to other facts.
Parents usually relax when they see the fence, because it answers the real question underneath the email: "will I be able to tell what my kid actually did?" Yes. That is the design (Class C).
The working-hypothesis posture
Some parents will ask a harder question: "Is this AI going to think for my kid?" The honest answer is a posture, not a promise.
We treat the underlying science, active inference and predictive processing in the tradition of Parr, Pezzulo, and Friston (2022), as a working hypothesis with growing, evidence-classed evidence, tested in the open (Class E). We do not tell students the science is settled. We tell them where the evidence is strong, where it is thin, and where a careful experiment would move the needle.
In the classroom that translates into a habit: predict, then check. Say what you expect. Say why. Look at what happened. Update in writing. A student who does that for a semester is harder to fool than a student who was told to trust or distrust "the AI." Distrust and trust are both shortcuts. Prediction and check are the work.
What we do not claim
We do not claim the tools in this classroom are intelligent in the way a person is intelligent. We do not claim they replace a teacher. We do not claim the science is finished. We do not use marketing language about the classroom on the school's public pages, because the school's job is to teach children, not to promote a vendor.
When a student asks whether the assistant "knows" something, the teacher answer is: it produces text that is often useful and sometimes wrong, and we are going to check.
If the parent wants more
Offer three things, in this order: a copy of the fence used in the class, a sample student artifact with the fence attached, and a short reading list of the sources the class actually draws on. Most parents stop after the fence. The ones who read further usually become the class's best allies.