How the AI works, and where it fails

What the AI is

The tutor is a computer program built on AI models made by Anthropic. It is not a person, it says so if you ask, and every screen that shows it calls it an AI tutor (or, when you teach it, an AI student). Each tutor has a name and a style, but every tutor is an AI and says so. It teaches from your own material (or the cited textbook passages of a starter topic), asks you questions, and checks your answers against a hidden answer key. For a College subjects outline taught as an AI explanation there is no material: it teaches from the AI's general knowledge, says on every screen that the lesson is not cited, and checks your answers against an answer key the AI wrote, which can be wrong.

When you teach the AI

In Teach mode (the "Teach" button with your buddy's name) you explain a topic and your study buddy plays a classmate who hasn't learned it yet. This is role-play by an AI: its first line says it is an AI playing a classmate, and it never claims to be a person, never says it has feelings and never asks about you. It only asks questions about what you said, one at a time, and never tells you whether you are right. Once in a session it may say a common wrong idea as if it believed it, to see whether you catch it; the report says whether you did. Our server checks every reply so the buddy does not give you the answer key's wording.

Teach is practice. It is graded against the same answer key as a lesson, but only the Check decides whether you have passed a topic.

What is sent, and when

Only when you study: the parts of your material a step needs, what you said, wrote or sent, and the answer key for that topic go to Anthropic to produce one reply or grade. The answer key is built and kept on our server. The full key is never sent to your browser while you study (you see one answer only when you choose to reveal it), and your data export lists your keys but not what they say, so they can't be used to look up answers. We do not use your notes, recordings or transcripts to train AI models. Anthropic deletes what we send within 30 days, except content its safety systems flag, which it may keep for up to 2 years.

How grades and levels are decided

The AI grades each answer against the answer key. The app itself, not the AI, decides the level of the next question, your mastery percentage and whether you passed (80% and a correct answer at the apply or transfer level). A revealed answer never counts toward the 80%.

A practice test is graded the same way, against each topic's answer key, and the app adds up the score; it never changes a topic's mastery. Flashcards are written by the AI from your material (or, for a College subjects outline, from general knowledge) and can be wrong: each card names the page it comes from where there is one, and you can remove a card you find wrong. Flashcards never change your mastery or your review plan.

Where it can be wrong

AI can misread an answer, miss a point or state something incorrectly. Check anything important against your course material and your instructors. If a textbook passage disagrees with your notes, the report says so instead of guessing which is right. An AI explanation has no source to check against, so it is more likely to be wrong than a lesson built from your own material or a cited pack. Check it against your notes or a textbook.

Safety

If you mention self-harm, abuse or a crisis, the tutor stops the lesson and points you to 988, a trusted adult and 911. It is not a therapist and does not diagnose. The Safety page is our published protocol and explains exactly what happens.

Tell us when it is wrong

Rate a grade after any session, or email support@sancyai.com. We read every report and use them to check the grader.