AI ethics and data

What the AI does here, what it sees, and what it doesn't · Last updated September 22, 2026

The short version

An AI model turns a teacher's plain-language description into a working app. That is the product. The model is Anthropic's Claude, reached through Anthropic's API; nothing we send it is used to train models; it never sees a student's name because the product holds none; and a person — the teacher — decides what gets built, what gets kept and what gets published. The model suggests and writes; it does not decide anything about a person.

Where the AI is used

• Building an app, in five steps a teacher can watch: planning (turning the description into a brief), clarification (asking what is still open), a library check (offering an app that already exists rather than building it again), a personal-information scan, and the generation itself.
• Revising an app a teacher already owns, from a plain-language change.
• Tagging a finished app so the library can be browsed.
• Suggesting ideas from a topic, a class's lesson notes, or an uploaded reference material.
• Sophia, the in-app assistant, which answers questions about a school or workspace using the same access the signed-in person has.

What is sent to the model

The description of the app and any follow-up answers; for a revision, the current app; for tagging, the app's title and description; for ideas, the topic, the titles of apps the school already has, recent lesson notes for a class the teacher owns, and the text of a reference material they uploaded; for Sophia, the conversation and the results of the lookups she makes on the person's behalf. Every one of these is something the signed-in person wrote or already has access to.

Not sent: student nicknames, staff email addresses, passwords or their hashes, billing details, or anything from another school. There is no roster and no student profile to send.

What the model provider does with it

Content sent through Anthropic's API is not used to train Anthropic's models. [State Anthropic's API data-retention period and whether we have zero-retention terms, once confirmed against the current agreement.]

Safeguards built into the product

• A personal-information screen, before anything is built. What a person wrote is checked for names, addresses, phone numbers and anything else that identifies a real person. A match stops the build and asks for a reword; the reason shown never repeats the information; an unclear answer blocks rather than passes.
• A person in the loop, at every step that matters. The teacher writes the description, answers the clarifying questions, sees what is about to be built, can cancel, reviews the result, and chooses whether to publish it. Nothing goes to students without a teacher having put it there.
• Nothing about a person is decided by the model. It never grades, ranks, flags or profiles a student, and it is not used for admissions, discipline, or any decision with consequences for a person.
• Costs are visible. Every model call is metered and shown in the school's billing, so nobody is surprised by what the AI was asked to do on their behalf.
• Every instruction we give the model is on file. The prompts that drive each step are plain-text files in the codebase, reviewable by anyone who asks.

Limits, honestly

The model makes mistakes. An app can contain a factual error, a bug, or a design a teacher would not have chosen. That is why the result goes to the teacher first, why revising it is a sentence away, and why nothing reaches a student until a teacher has put it there. We do not claim the model's output is correct; we claim the teacher can see it and change it before anyone else does.

The personal-information screen is a model's judgement too, and it can miss things. It is a safeguard, not a guarantee, which is why the product also collects so little that could be harmful to miss. [The screen does not currently run on the join-link flow a student uses from a class code. Either close that gap or say so plainly here before publishing.]

Your data, the model's outputs

What you write stays yours. What the model writes for you is yours to keep, change, publish or delete, as the Terms of Service set out. We do not use your descriptions, apps or notes to build anything for anyone else, and we do not use them to train models of our own — we have none.

See also: Privacy Policy · COPPA · FERPA · Questions: [privacy contact address]