AI Policy

• Atlas, the AI tutor inside every lesson: it answers your questions, explains ideas in different ways and helps you practise.
• Practice: grading open answers and giving feedback, and suggesting what to review next on your agenda.
• Tutor memory: remembering things you tell Atlas — your goals, preferences and earlier mistakes — so later sessions fit you (you control this; see section 5).
• Course content: our courses, lessons, quizzes and flashcards are drafted with AI, as are the articles and explainers on our site.
• Lesson audio: lessons can be read aloud by a computer-generated voice.
• Help: some answers in the Help Center are drafted by AI from our help articles.

Atlas uses large language models from Anthropic (Claude) and OpenAI, hosted for us by Amazon Web Services in Amazon Bedrock. Which model answers depends on your plan and on availability. If Bedrock cannot answer, the request may go through a back-up route, OpenRouter, to a model from the same companies; those requests are sent with the provider's data collection turned off. The current list of providers is on our Sub-processors page.

Atlas is not a general-purpose chatbot. Each answer is built around the course and lesson you are studying, and it is shaped by what you have told us about yourself so that examples and pace fit you.

To answer a question, Atlas sends the model:

• your question, and the recent messages of the conversation (or a short summary of earlier ones);
• the relevant parts of the course and lesson you are on;
• the learner profile you confirmed — such as your background, goal, experience and how you like things explained;
• a short summary of your progress, such as which ideas you have mastered and which need work;
• memories Atlas has saved, if tutor memory is on;
• excerpts of study material you uploaded, when they are relevant to the question.

Before your question is sent, common identifiers — email addresses, phone numbers, card numbers, government ID numbers and similar — are removed from it. We do not send the model your email address, your payment details or your password. Please do not type sensitive personal information into the tutor that the lesson does not need.

• Tutor conversations (your messages and Atlas's answers) are saved to your account so you can pick up where you left off. They are kept for as long as you have your account and are deleted when you delete it.
• Tutor memories are kept until you delete them in Settings › Tutor memory, or delete your account.
• Your feedback on an answer (thumbs up or down) is stored with that answer.
• Records of how many tutor turns you used, which we need for plan limits, are kept for up to one year.
• Material you upload for the tutor is kept for as long as you have your account.

To grade open answers consistently, we keep a store of answers and the grades they received. Answers in this store are stripped of common identifiers and are not linked to your account, so it is used to grade similar answers from other learners and cannot be traced back to you.

How account deletion works, and how long backups last, is described in our Privacy Policy and on our Security page.

• Turn tutor memory on or off with "Let Atlas remember" in Settings › Tutor memory. With memory off, Atlas does not save new memories or use saved ones in its answers. Your conversations are still saved to your account history.
• See every memory Atlas has saved — what kind it is, when it was saved and which conversation it came from — and delete any of them in Settings › Tutor memory. A deleted memory is removed straight away.
• Download a copy of your data from Settings › Privacy.
• Delete your account, including your tutor conversations, in Settings › Security.

We do not use your tutor conversations, answers, notes, uploads or other learning data to train or fine-tune AI models, and the AI providers we use are not permitted to use them to train theirs. When we improve the models we use to write courses, we use our own course material only.

• Courses are drafted with AI, checked automatically against our content standards, and reviewed by our team before they are published. Drafts that fail the checks go to a person.
• Our staff tools do not show tutor conversations. Team administrators never see them either: on a team plan, admins see learning progress only — not tutor conversations, answers or notes. Access to the underlying data is limited to the engineers who run the service and is used only to operate, secure and fix it, or when the law requires.
• When you report a problem with an answer, or send us a help request, a person reads it.

The models we use have their own safety systems. If one of them stops an answer, Atlas tells you so rather than showing a partial answer as if it were complete. We also separate your input from Atlas's instructions to make it harder to manipulate, and apply usage limits — your plan's monthly tutor turns — to keep the service fair and available. Misusing the tutor is against our Acceptable Use Policy.

AI can be confidently wrong. Answers, feedback, grades and AI-written content may contain mistakes or be out of date. They are for learning, not professional, legal, medical, financial or career advice. Check anything important with a reliable source, and tell us when Atlas gets something wrong — use the thumbs-down on the answer or Help › Report a bug.

Our AI features do not make decisions about you that have legal or similarly significant effects. Grades and suggestions shape your study plan and nothing else.

AI moves quickly, and we will update this page when the way we use it changes — for example a new kind of AI provider or a new use of your data. The date and version at the top show when it last changed. Questions: privacy@educationpals.ai.