How to Build an AI Detection Policy That Doesn't Punish Your Best Students

The Problem a Blanket Ban Creates

A ban communicates that AI use is shameful, hidden, and unmentionable. That is exactly the wrong message for a generation that will use these tools in every professional environment they enter. It also creates a practical enforcement problem: a rule the teacher cannot enforce, in a room full of students with smartphones, is not a rule. It is a wish.

The more important issue is what happens to trust. A student who uses AI in good faith on an assignment that didn't clearly prohibit it, then gets flagged, has a legitimate grievance. A student who knows the tool exists and chooses not to use it on a Red assignment because the rule was clear has made a meaningful decision. The policy is what makes the second student possible.

Nearly 90% of students are already using AI in some form. When Securly analysed actual student network usage data in March 2026, they found that when districts set clear guardrails, 80% of student AI interactions fell within approved parameters. Students follow clear rules. They fill ambiguity with whatever serves them.

What Disclosure-Based Policy Actually Means

A disclosure-based policy is not a free-for-all. It requires students to state what AI they used, how they used it, and what they did with the output. The grade reflects the student's critical thinking on top of the AI's work. Not the AI's work alone.

The disclosure log is also evidence. If a student claims they didn't use AI and their disclosure log is blank, that is a verifiable fact. If a student says they used AI for brainstorming but the version history shows the entire draft was pasted in a single event, that is a policy violation with documentation. The disclosure requirement shifts the evidentiary burden back to the student in a way that a detection score never does.

Disclosure-based policy also gives teachers visibility into how students are actually using AI. A teacher who discovers that 80% of the class used AI for the outlining stage on a Yellow assignment can adjust future instruction accordingly. A teacher whose students are hiding their use learns nothing.

The Traffic Light Framework

Three colours. Three clear rules. One designation per assignment, printed on the rubric before the student begins.

Green means AI is permitted at all stages. Brainstorming, drafting, editing, formatting. The student must submit a disclosure log stating what tool they used, which parts of the work it touched, and what they did to revise or extend that output. The grade reflects their critical thinking on top of the AI's first pass.

Yellow means AI is permitted for ideation, outlining, and grammar checking only. All writing is original. No AI-drafted text in the submission. A process portfolio is submitted alongside the final draft showing the outline, the brainstorm notes, and the version history link.

Red means no AI at all. In-class timed writing, oral presentations, lab reports completed under supervised conditions, device-free assessments. When the environment is controlled, the policy enforces itself.

The framework works because it is specific per assignment, not a global position. A Green project in Term 1 teaches students to use AI responsibly. A Red exam in Term 2 assesses whether they can produce independent work under pressure. Both are legitimate instructional goals. Neither requires a detector.

How to Write Green, Yellow, and Red Assignments

A Green assignment is well-suited for research projects, multi-draft essays, and any task where the thinking process matters more than the final document. The requirement is a disclosure log and a process portfolio: notes, drafts, version history link. Students who did not use AI submit the same portfolio. The absence of AI use in the log is a legitimate and accepted disclosure.

A Yellow assignment suits tasks where you want the student to do all the writing but you are realistic that they may use AI for outlining and structure. An analytical essay where the argument must be original is a Yellow task. The rubric specifies that AI-drafted sentences in the body of the submission constitute a policy violation, regardless of whether a detector flags them.

Red assignments are where the evidence of independent work is the point of the task. Timed in-class writing. The live oral presentation. The experiment that requires the student to be physically present at a bench. Any assessment where the process is the product. These require no detection and no disclosure. The conditions do the work.

What to Put in Your Syllabus Right Now

Three paragraphs cover the essentials. Here is the structure with the language that matters most.

First paragraph: the general AI position. "AI tools including ChatGPT, Claude, and Gemini are permitted in this class on designated Green assignments with a required disclosure log. They are restricted to ideation and grammar checking on Yellow assignments. They are prohibited on Red assignments, which are completed under in-class conditions."

Second paragraph: the disclosure requirement. "Any use of AI at any stage of a Green or Yellow assignment must be disclosed in the submission. The disclosure log states the tool used, the task it assisted with, and how the student revised or extended that output. Submissions without a disclosure log on Green assignments will not be graded until one is provided."

Third paragraph: the consequence statement. This is the one that protects the teacher legally. "An AI detection score is not sufficient evidence to initiate a grade penalty or academic misconduct finding. Where AI use is suspected outside the terms of this policy, I will request the document version history and schedule a brief verbal check with the student before taking any action."

That third paragraph is the most important one. It removes the detection score from the evidentiary chain and replaces it with observable process evidence. It also signals to students that a false flag will not result in an automatic penalty, which reduces the anxiety that drives students to degrade their own writing to avoid detection.

How to Handle the Cases the Policy Doesn't Cover

Three scenarios come up repeatedly. Each one has a single next step.

A student submits a Green assignment with no disclosure log. The submission goes back ungraded with a request for the log. No accusation. No grade penalty until the process requirement is met. This is administrative, not adversarial.

A Red assignment gets flagged by a detector. The teacher requests the version history and schedules a two-minute verbal check. The student explains their thesis and identifies two sources. If they can do that, the case is closed regardless of what the detector said. If they cannot, the version history is the evidence, not the score.

A detection score contradicts the teacher's own judgment. The teacher's judgment is more reliable than a probabilistic classifier for students they know. A student who has been writing consistently at a particular level all term and produces work that matches that level is not a detection case. Document the decision and move on. The process verification steps are the evidence structure when a finding does need to be documented formally.

Traffic Light Syllabus Generator

Select the AI functions you want to permit in your class. The tool generates a copy-pasteable policy paragraph for your syllabus.

FAQ