A student turns in an essay. A teacher grades it. Somewhere between those two moments, a question sits unanswered: who, or what, actually did the work?
The AI-authorship fence is a classroom policy that puts that question in writing on every artifact. Not once a semester. Every artifact. It is a small piece of metadata that names the human author, names any tool assistance, and describes what the tool did. Students learn to declare. Teachers model the same discipline on their own materials. The fence is not a punishment or a suspicion device. It is a reading aid, a thinking aid, and a defense against a market that is currently selling schools a great deal of vapor.
What the fence actually is
An authorship fence is a short, structured block attached to a piece of student or teacher work. It answers four questions in plain language.
- Who is the human author of record.
- Which tools, if any, assisted, named specifically (a spellchecker, a graphing calculator, a large-language-model chat interface, a translation service, a citation formatter).
- What each tool did, in a verb: drafted, outlined, corrected, translated, formatted, searched.
- What the human did with the tool output: rewrote, kept verbatim, discarded, verified against a source.
That is the whole shape. It fits at the top or bottom of a document, in a footer on a slide, in the notes on a lab report. It is boring on purpose. Boring metadata is metadata that gets filled in (Class C, configuration/integration: this mirrors provenance conventions in scientific publishing and in software supply-chain manifests, which work because they are dull and mechanical).
Why the fence protects the learner
The pressure students feel to hide tool assistance is real and it does damage. When hiding is the default, three things happen. The student loses the chance to reflect on what they actually did. The teacher loses the ability to teach the difference between a tool's suggestion and the student's judgment. And the artifact itself becomes uninterpretable later, because no one remembers which sentences the student wrote and which the tool wrote (Class F, falsifier: if the fence policy is in place and a random sample of graded artifacts shows no meaningful change in what students report doing, the policy is failing and we should say so).
The fence flips the incentive. Declaring is normal. Declaring is expected. A student who used a chatbot to outline a five-paragraph essay writes that down, and then the conversation in class is about the outline: was it good, what did you change, what would you change now. The teacher gets a lever. The student gets practice at the thing school is supposed to teach, which is judgment.
Why teachers must model it too
A policy that applies only to students is a policy about student behavior. That is not what this is. Teachers use tools. Teachers use lesson-plan generators, quiz builders, image search, translation, grammar checkers, and, increasingly, chat interfaces to draft rubrics and worksheets. If teachers do not fence their own artifacts, the message to students is that the fence is a compliance ritual for the powerless. It becomes theater.
Modeled honestly, the fence is contagious. A teacher who writes at the bottom of a handout "drafted with LLM chat, revised by hand, examples replaced with local ones" teaches more about intellectual honesty in one line than a semester of academic-integrity slides. The teacher is not confessing. The teacher is describing the work, the way a chemist describes reagents.
What the fence is not
The fence is not a lie detector. It is not a plagiarism gate. It is not an AI-detector, and it should not pretend to be one. AI-detector software has a poor track record and tends to punish second-language writers and neurodivergent writers most (Class U, unverified in our own data but repeatedly reported in the literature; we cite it here as an open concern, not a settled fact). The fence works because it is a norm, not a scanner. It presumes good faith and rewards declaring.
The fence is also not a claim about the tool's intelligence. It does not say the tool "wrote" in any authorial sense. It records that the tool produced text, image, code, or translation that the human then did something with. That distinction matters, because the vendor language pushing into schools right now blurs it deliberately.
The contrast with vendor slogans
The current market for classroom software is loud with sweeping adjectives, world-changing this and reinvented that, personalization at scale, and so on. Set the adjectives aside. Read a vendor page slowly and ask two questions. What does the tool actually do to a specific artifact in a specific lesson. And what would count as evidence that it worked or failed. If the page cannot answer either, the tool is not ready for a classroom. The fence gives you a way to hold the line: whatever the tool did, it goes on the artifact. If the tool did something a teacher would be uncomfortable seeing named, the tool should not be in the room.
This is where our public frame matters. Our own program, Universal Natural Intelligence, is a working hypothesis on an attainable path toward General Natural Intelligence, natural not artificial. We are not selling a classroom AI. We are teaching a discipline of transparent tool use, evidence-classed claims, and honest declaration. The fence is the classroom-scale version of the same discipline we hold ourselves to on our own research pages.
For field-level context on why the framing shift matters, see Themesis, The AGI Landscape Just Changed; in our reading, this is a signal that the ground under vendor claims is moving and schools should not be locked into last year's marketing.
How to start on Monday
You do not need a district rollout. Pick one class. Add a four-line fence template to the header of every assignment. Fill it in on the assignment sheet you hand out, because you built the sheet with tools too. Ask students to fill it in on what they turn in. For the opening two weeks, make the fence itself the thing you grade honestly: is it filled in, is it specific, is it plausible. After that, the fence recedes into the background and the conversation moves up a level, to what the student thought.
Expect friction. Some students will underreport, out of habit or fear. Name that gently and keep going. Some parents will ask what this is. Show them the template. Some administrators will want to bolt on a detector. Decline, and explain why the norm is stronger than the scanner (Class E, expert citation: this is consistent with what integrity researchers such as Tricia Bertram Gallant have argued for years, that culture beats surveillance in academic-honesty work).
The transparency spine
The fence is the anchor vertebra. It supports everything else this site will teach: how to audit a classroom tool before adopting it, how to teach the science of learning without smuggling in vendor claims, how to hold the line between using a tool and being used by one. It is small on purpose. Small policies that everyone actually follows are how a school changes, quietly, without a rebrand.