How Fairgrade uses AI

Last updated: 1 October 2026

The short answer

Fairgrade is not an AI grader. Grades on Fairgrade come from human judgement: your peers' reviews, weighted by how credible each reviewer has shown themselves to be, and finalised by your instructor. That weighting is done by the Marciano Method, which is mathematics grounded in probability theory, not a machine-learning model, and it explains every weight it assigns.

AI has a place beside that. It is a second opinion, labelled as such, that an instructor may choose to see. It never decides a grade, and it never learns from your work.

Where AI is used today

Feature	Who turns it on	What is sent to the AI provider	What comes back	Who sees it
AI scoring (optional second opinion)	The instructor, per assignment	The submission (file or text), the rubric, the assignment brief and any reference material the instructor attached	A score per rubric criterion, a confidence level, and written feedback with the model's reasoning, all labelled as AI	The instructor always. Students, only if the instructor chooses to show AI scores or AI feedback when publishing results
Summary of peer feedback	Runs only when the instructor chooses to show the feedback summary when publishing results	The written peer reviews on one submission	A short structured summary of what reviewers said	The student, with their results
Assignment readiness check	The instructor, when uploading an assignment or rubric	The instructor's own document (the brief or the rubric); no student data	Whether the document has enough substance to grade against, with a short reason	The instructor only

What is never sent: your name, your email address, your student number, your login identifier, or anything else that identifies you. If you write your name inside your work, it travels with the work; when anonymous grading is on, Fairgrade warns you before that happens.

Which AI: Anthropic's Claude models, in tiers the instructor picks (faster or deeper), run through Amazon Bedrock inside Fairgrade's own AWS account in the United States. Anthropic never receives the data; Amazon Bedrock does not store it and does not use it to train models. See our subprocessor page at /subprocessors.

What AI is not used for

Detecting AI use. Fairgrade does not try to detect whether a student used AI, and does not police it. Detection is unreliable and punishes honest students with false positives. What Fairgrade measures instead is the quality of a student's judgement in reviewing others, which cannot be outsourced without leaving a trace.
Deciding grades. No grade is final until an instructor makes it so.
Weighting reviewers. Credibility is estimated by the Marciano Method from the record of reviews, not by an AI model.
Checking for names. The anonymity check runs on our own servers with no AI involved.
Anything about you as a person. AI never receives who you are.

For instructors

AI scoring is off unless you turn it on for an assignment. When it is on, every AI score and comment is labelled, and you decide whether students see them.
AI output can be wrong, incomplete or biased. Treat it as one more reviewer of unknown reliability whose opinion you weigh, not as a verdict.
Using AI features spends AI credits, shown on the assignment page.

For students

You can see on your results page whether AI was used, and what it said, when your instructor has chosen to show it.
Whether and how you may use AI in producing your own work is your instructor's rule, not Fairgrade's. Read the assignment brief.
Your data rights run through your institution. Questions about AI on Fairgrade: privacy@fairgrade.com.

When this page changes

Every AI feature is listed here before it is available. A feature that sends the same kinds of data as those above is added here with a date. A feature that would send a new kind of data, or identify a person to an AI provider, would require a new version of the Privacy Notice with 30 days' notice first.

Date	Change
1 October 2026	First published.