What responsible AI use
looks like in one real course.

Schools, programmes, and First Nation administrations across Canada are being offered AI tools with no shared understanding of what responsible use looks like in practice. The default risk is familiar: software handed to communities by people who do not have to live with the consequences.

This case study is a concrete answer to "what does responsible look like." It uses a real student doing real coursework, with the AI audit logs and source files attached, so any educator or administrator can see what was done and adopt the parts that fit.

The course was delivered through a partnership programme between a First Nation and a public school board. Across the unit, the student submitted five journals, one home energy audit, and a culminating environmental careers project. Each submission was paired with an AI Use Audit Log describing how AI contributed and what the student decided.

Every assignment began the same way. The AI built a helper page summarising the unit's source material, pre-filled the rows where defaults were sensible, and flagged anything that was not from the slides as interpretation. The student picked the prompt or angle. The AI then asked one guiding question at a time, drafted a response using the student's actual answers, and waited for approval before moving to the next.

The student decided the prompt, the personal examples, the corrections, the closing argument, and what made the cut. The AI carried the structure, the source material, and the formatting load.

The standard "I used AI to help me write this" disclosure is procedural and weak. The audit logs in this work go further. Every entry records the student's input, the AI's contribution, the sources used, the verification step, and any remaining gaps. A separate "Knowledge Transfer" section captures the actual facts surfaced during the session, with the sources that verified them.

This is the difference between disclosure and accountability. A reader can reconstruct what the student learned, where the facts came from, and which judgements were the student's own.

Corrections

One audit log documents the student rejecting an AI overclaim about a workplace hazard mid-walkthrough. The draft was revised on the spot and the correction is in the record.

Source separation

One unit briefing labelled every fact as either slide quote, slide fact, AI reading, or general knowledge. The final journal cited accordingly. When a unit had no posted slideshow, the journal did not invent "the slides say."

Verified facts

Salary ranges, government statistics, and provincial grid data were fact-checked against named authoritative sources, including Natural Resources Canada, IESO, CBC News, Hydro One, ZipRecruiter, and Glassdoor, before any of it reached the draft.

Knowledge transfer

Each log lists facts learned during the session, with the source for each. A reader who has never seen the original course materials can still trace where every claim originated.

AI did not pick the student's prompt. It did not generate the firsthand mining-industry recollections that anchored several journals. It did not catch its own formatting or overclaims without the student saying something. It did not decide what the student's opinion was on any of the contested questions in the course.

None of that is a weakness of the pattern. It is the design. The AI is a tool with a defined job. The student is the author. The audit log is the receipt.

Audit log template

Date, student input, AI assistance, sources used, approval, remaining gaps. Plus a knowledge transfer section.

Helper page per assignment

One HTML page that holds the unit notes, the prompts, the guiding questions, and a draft area. Students keep their work in one place.

One question at a time

The AI asks the next thing it needs. It does not generate complete answers up front. The student stays in control of the draft.

Fact-check before finalising

Any numeric or factual claim in the draft is verified against a named source before submission. Sources are listed in the audit log.

Source separation

Treat slide facts, AI interpretation, and general knowledge as three different things. Label each one in the briefing the AI gives the student.

OIATC's premise is that the systems meant to serve First Nations are rarely chosen by them. AI in education is a current edge of that problem. A Nation rolling out ChatGPT or a similar tool without a governance framework is doing what we have seen done with case management, data platforms, and grant systems. Adopting the tooling without the protocol that gives it shape.

The pattern documented here is not the framework. It is a working example of what one piece of the framework can look like at the classroom level. Protocol can scale up from this. Product without protocol cannot.