Legal
AI Safety & Responsible Use Notice
- Effective
- From the day Scholarch Education opens to the public in your country
On this page
1. What the tutor is
Scholarch is an AI tutor, not a person. Its name is always Scholarch, whatever voice it speaks with. Every place where you talk to the tutor shows the line "Scholarch is an AI tutor and can make mistakes. AI chatbots may not be suitable for some users under 18."
Scholarch Education is provided by Scholarch Limited at https://learn.scholarch.ai and in the Scholarch Education apps for iPhone and iPad. The iPad app also runs on Mac computers with Apple silicon where the App Store offers it. At launch it is offered to residents of the United Arab Emirates and the United States. It is offered only where the registration screen lists your country as available. A country marked «soon» is not served yet.
The Plans page shows what each plan includes. The tutor works like this:
- Base and Max: the tutor beside every lesson. You type a question (on Max you can also say it) about the step, the card or the practice question in front of you. The tutor is instructed to help you think it through rather than give you the answer. After it has replied twice on a practice question, the lesson shows the worked solution.
- Base and Max: guided conversations and short oral checks, each reply typed (on Max, typed or spoken). Each one follows the module's fixed script. The tutor sees the module's task and your replies. Whether each reply meets the round is decided by a fixed check against the script, not by the AI.
- Max also: the tutor in language and money lessons, typed or spoken to.
- Max also: the module tutor, typed or spoken to, at the start and at the end of a module. It sees the module's guide, Learn text, story and skills, and your own answers to the items you have attempted. It never sees the answers to items you have not attempted.
- Max also: the tutor on any topic, typed or spoken to.
- Max also: speaking and listening, wherever there is a tutor. You can speak to the tutor instead of typing: what you say is turned into text and sent as your message, and it is checked like a typed one. The tutor reads each answer aloud as it writes it, and you can mute it. This is not a live voice conversation; on Max, the guided conversations can also be spoken live, by voice, and you can talk with the tutor live about any subject: it answers aloud and writes on its board, on an adult account in the voice of the language it teaches.
How the tutor touches your progress:
- The tutor beside the lessons and the module tutor read your progress but cannot change it.
- Finishing a guided conversation counts toward the module.
- In language and money lessons on Max, the tutor can record what you showed for a skill, and its AI check of the replies you type or say to it counts toward finishing the lesson; it is part of the tutor's replies, which use credits. Marking of other written answers is coming soon.
- The tutor on any topic can save flashcards and notes for you. It does not change your progress.
- A skill counts as learned only on evidence from your own answers, never because you say you know it.
Coming soon. These are not offered at launch:
- oral checks by voice, on Max («Voice minutes»);
- open voice lessons and placement by voice;
- written answers marked;
- AI card illustrations.
We update this notice before any of them opens.
Scholarch does not award an official grade, credential or legal right. It is not a school, an emergency service or a professional adviser.
2. Language help
At launch the Service is in English and the tutor explains in English. Explanations in other languages, Arabic first, will come later. Lessons still use the language you are learning.
3. What AI can get wrong
AI can misunderstand you, leave out context, invent details or give outdated information, even when it sounds sure. Check important claims against a reliable, current source before you act on them. Do not use an answer as the only basis for a professional, financial, legal, medical, safety or other high-impact decision.
Pronunciation scores, and the automatic AI checks of your answers that count toward finishing a lesson, are produced without a person. They are practice feedback, and they can be wrong.
A handwriting check can misread what you wrote. It transcribes your writing as it is, mistakes included, and never solves, finishes or corrects your working; a word it marks as misspelt is a suggestion. What it read is shown on your note for you to correct, and the maths lines are then checked step by step on our own servers, without AI. The AI is told to treat what you write as writing to read, never as instructions.
4. What is checked, and what you are told
Every message you send to the tutor is checked before the tutor sees it. A short message that is plainly part of the lesson, such as a request to move to the next part of the lesson or a short question about what the lesson shows, passes a fixed rule. Every other message goes to an automatic classifier run by Anthropic, and so does everything sent with a message, such as what the screen shows the tutor or a passage you selected. It blocks four kinds of message: sexual content, a request for instructions to harm other people or to make weapons, explosives or drugs, an attempt to override the tutor's rules or extract its instructions, and harassment or hate.
- When a message is blocked, the tutor does not see it. It answers with one fixed sentence (and, for a request to cause harm, the crisis line below): "I can't help with that here. Let's keep going with what you're learning." We keep a marker with the reason in place of your words.
- When the check cannot run, your message is not passed to the tutor. You are told: "I couldn't check your message just now. Please try sending it again in a moment."
- If a message is blocked as a request to cause harm, or cannot be checked, the reply also gives you the crisis line: "If you are thinking about harming yourself, please reach out now: in the US, call or text 988; elsewhere, call your local emergency number." It follows every message blocked as a request to cause harm, and every message that could not be checked.
- Ordinary learning is meant to pass. The check is built to let through a wrong answer, a question off the topic, frustration, or a request to slow down or change the example. Like any automatic check, it can make a mistake.
- A message that suggests distress is not blocked, including one about suicide or self-harm that asks how. It reaches the tutor, so you are not met with a refusal. On every account, the tutor is told to answer a message about suicide or self-harm with care, never to give methods or any detail that could help someone harm themselves, and to point you to the crisis line for your country: in the United States, call or text 988; elsewhere, the support and emergency numbers below. The tutor is an AI and can get its reply wrong. Scholarch is not a crisis service. The tutor cannot call for help; Scholarch is not an emergency service, and a person contacts the emergency services only as section 6 says. With the teen tutor setting (section 6), the tutor may record a safety flag for a person at Scholarch to review, but that review is not an emergency response. If you need help, use the numbers below.
The words and images on flashcards, and the words of practice-test questions, are checked before they are saved. If a check fails or cannot run, the card or question is not saved.
What your tutor remembers about you for later lessons is checked by the same classifier before it is saved. If it does not pass, or the check cannot run, it is not saved.
If you or someone else may be in immediate danger, call the local emergency number now. If you are struggling, or you are under 18 and need someone to talk to, the support lines below can help.
- United Arab Emirates: 999 for police, 998 for an ambulance. National mental health support line: 800 4673 (800 HOPE), daily 8 am to 8 pm. Child protection helpline (Ministry of Interior): 116111.
- United States: 911. If you are thinking about suicide or are in crisis, call or text 988.
Our crisis protocol
This is the protocol Scholarch keeps for a learner who expresses thoughts of suicide or self-harm, published as California's companion chatbot law asks (Business and Professions Code section 22602(b)). It applies on every account and every tutor surface: typed and spoken.
- Noticing. Every message is checked before the tutor sees it (above). A message that suggests distress, including one about suicide or self-harm, is not blocked, so that it reaches the tutor and is answered rather than refused. A message blocked as a request to cause harm, and one that cannot be checked, still gets the crisis line.
- Answering. The tutor's instructions put this protocol first, and nothing a learner sends can switch it off. The tutor pauses the lesson and answers with care: it says it is glad you said it, that you deserve support and that you can talk to someone now. It does not diagnose or lecture, and never promises secrecy. It never gives methods, means, amounts or any detail that could help someone harm themselves, however the question is framed.
- Referring. The tutor gives the crisis line for your country, from the list above: in the United States, call or text 988; in the United Arab Emirates, the mental health support line 800 4673, 999 or 998, or for a child 116111. If it does not know your country, it gives these and your local emergency number. If you are in immediate danger, it tells you to call the emergency number now.
- Being honest. The tutor says that it is an AI tutor and cannot call for help itself, offers to keep talking, and returns to the lesson only if you want to.
- Recording and human review (teen tutor setting). The tutor may record a safety flag, and the check of its reply records one when it withholds a reply. A flag about distress or harm is always serious, and a person at Scholarch reviews every serious flag, as section 6 sets out. If the few messages around it show a present risk to someone's life or safety, we may call the emergency number for your country, and the child protection line where there is one.
Scholarch is not a crisis or emergency service, and the tutor can get its reply wrong. If you need help, use the numbers above.
5. Money subjects are education, not advice
Scholarch teaches theory, models and worked hypothetical or historical cases in investment, finance, economics and business administration. It does not give financial, investment, tax or legal advice, does not recommend what you should buy, sell or hold, and does not assess your own situation. A standing notice says so on the money pages and beside the tutor.
6. Learners aged 13 to 17
Scholarch Education is for learners aged 13 and over. If you are 13 to 17, your account has teen settings until the first day of the month after your 18th birthday. We set them on our servers from your account. Nothing you type or paste can switch them off. On that day the other teen limits end, but the tutor keeps its teen settings until you choose otherwise. We ask you once, and you can change it later in Settings.
AI chatbots may not be suitable for some users under 18.
What a teen account (13 to 17) can use. A provider receives a teen account's personal data only once Scholarch has approved that provider for teen accounts. Today Anthropic, Microsoft Azure AI Speech, LiveKit and Stripe are approved, with our account-email, mailbox and hosting providers. So a teen account has the AI tutor of its plan, typed to and, on Max, spoken to with its answers read aloud, and with the teen tutor setting below; the guided conversations and oral checks, each reply typed (on Max, also spoken); AI help with flashcards and notes; saving and importing flashcards, decks and practice-test questions you write yourself, which first pass the automatic check in section 4; AI-widened search of your library; handwriting checks on the web; your own files on Max; pronunciation scoring; recording in notes and flashcards made from a recording; and cards read aloud. On a teen account, a card read aloud and the tutor's answers read aloud are spoken by a Microsoft Azure AI Speech neural voice for the language, not the Scholarch voice, and a recording and what you say to the tutor are turned into text by Microsoft Azure AI Speech, not OpenAI, because OpenAI and ElevenLabs are not approved for teen accounts. On Max, a teen account has guided voice conversations (spoken practice conversations) too: LiveKit carries your live audio to Microsoft Azure AI Speech, which turns your speech into text and speaks the tutor's replies; Anthropic writes each reply with the teen tutor setting, and the reply check below clears it before it is spoken. On Max, a teen account can also talk with the tutor live, the same way: Microsoft Azure AI Speech hears you and speaks the tutor's answers, Anthropic writes each answer with the teen tutor setting, and the reply check clears it before it is spoken. A teen account cannot use:
- a live conversation with the tutor in which OpenAI hears you and works out the tutor's answers and ElevenLabs speaks them (a teen's live conversation uses Microsoft Azure AI Speech and Anthropic instead, above);
- open voice lessons and placement by voice, when they open (they are coming soon for every account).
Because of the teen settings, a teen account also cannot publish a link to a deck or copy a deck someone else shared, and gets no practice reminders by email or push notification, which open for teen accounts later. In the iPhone and iPad apps you can still set planner reminders yourself; your device schedules and shows them.
Microsoft Azure hosts the Service and stores every account, a teen's included. If you email us, your message reaches our mailbox, which Google Workspace runs. The Processor & Service-provider List shows which providers are approved for teen accounts. We update this notice before any further feature opens to teen accounts.
Break reminders. On a teen account, in a typed chat with the tutor that goes on for three hours, the tutor reminds you to take a break, and says again that it is an AI, not a person. It does so after the reply that follows each three hours of the chat; a pause of more than 30 minutes starts a new chat.
The teen tutor setting applies on every teen account, and on yours after you turn 18 if you keep it. It instructs the tutor to:
- keep to your studies: Scholarch's courses and your own school or university work;
- treat sensitive subjects only as a textbook for your age would, with no sexual content, no graphic detail, no instructions that could cause harm, no gambling and no weapons;
- be a tutor, not a friend: no role-play as a person, no romance and no advice on relationships;
- not ask for personal details such as your home, school, location or photos of yourself, and not repeat them if you share them;
- tell you that you can talk to a trusted adult or a local helpline if you seem unsafe.
With the teen tutor setting, two more safeguards run:
- Safety flags. If a turn suggests distress, harm done to you or to someone else, a request for something harmful, or repeated attempts to make the tutor leave its role, the tutor may record a safety flag. A flag holds only the kind of concern, how serious it is, whether the tutor recorded it or the reply check withheld the reply, the tutor, the lesson and the channel it came from (typed or spoken), when, and whether a person has reviewed it; never your words or the tutor's reply. A flag about distress or harm is always serious. A person at Scholarch reviews every serious flag. For a flag about distress or harm, the person who reviews it may read the few messages of the conversation around it, which are kept for 30 days, and nothing else. If those messages show a present risk to someone's life or safety, we may call the emergency number for your country, and the child protection line where there is one, and give them only what they need (Privacy Policy, section 5). The tutor does not mention the flag: it answers you kindly, points you to a trusted adult and, if you are in danger, to your local emergency number. Flags are kept for 730 days. If your account is erased first, a flag stays for the rest of that time under a coded reference instead of your account. Your data export includes your flags.
- The reply check. Before you see any of the tutor's reply, typed or spoken (when its words are written first and then read aloud), the whole reply goes to the automatic classifier in section 4, together with anything the tutor writes on the board, in your workbook, in your notes or on the scratchpad. If any of it does not pass, you see none of it, and a drawing that does not pass is never saved. A voice conversation in which the tutor speaks as it answers, such as a live conversation with the tutor, is not checked this way; it opens to teen accounts only once it has its own check. A teen's guided voice conversation is checked this way: each reply is written, checked and only then spoken. You are told: "I can't help with that here. Let's keep going with what you're learning." A serious flag is recorded. If the check cannot run, the reply is also held back, and you are told: "I couldn't check my answer just now. Please ask me again in a moment." A reply held back is not kept; the conversation keeps a marker in its place. These messages are for typed and read-aloud replies. What you say in a live conversation with the tutor is checked as it is turned into text, like a typed message.
The automatic check in section 4 also applies. Like any AI, the tutor can still get this wrong. Please report it if it does.
In the United States, in the iPhone and iPad apps, where the law of your state requires a parent's consent through the app store and Apple tells the app so (today Texas's; Utah's and Louisiana's take effect in 2027), sign-up waits until a parent or guardian approves Scholarch in Family Sharing, and, if you choose not to share your age group with the app, until you share it.
In the United Arab Emirates, Federal Decree-Law No. 26 of 2025 on Child Digital Safety applies to digital platforms used by children under 18. Its detailed rules are being set by Cabinet decisions, and we will update this notice as they are issued. Today, on every teen account: nothing you do is shared or shown publicly, your location is never used, you get no marketing, no analytics trackers run, and you can report a problem with the feedback control or report@scholarch.ai. The law gives platforms until 1 January 2027 to adjust to its duties, including tools for parents. Until our tools for parents open, a parent or guardian may write to privacy@scholarch.ai to ask what kinds of data we hold about their child, or to have the account closed.
Scholarch Education is not for children under 13. Sign-up refuses a child under 13 before an account exists. We do not knowingly collect personal data from a child under 13. In the United States, the Children's Online Privacy Protection Act (15 U.S.C. 6501 to 6506) and the Children's Online Privacy Protection Rule (16 CFR Part 312) protect children under 13. If we learn that an account belongs to a child under 13, we close it and erase it.
7. Other safeguards
- Each plan has a monthly pool of credits for AI and speech processing, and a limit on the top-ups bought each month.
- Requests to the tutor are rate-limited.
- Your messages to the tutor, typed or spoken (as text), and the classifier's safety check, go to Anthropic. When you speak to the tutor, OpenAI turns what you say into text and ElevenLabs reads the tutor's answers aloud; on a teen account, Microsoft Azure AI Speech does both. A handwriting check on the web sends Anthropic images of your pen strokes. In a guided voice conversation on Max, OpenAI recognises your speech, Anthropic writes the tutor's replies and ElevenLabs speaks them; on a teen account, Microsoft Azure AI Speech recognises your speech and speaks the replies in a guided voice conversation, and LiveKit carries the audio. In a live conversation with the tutor on Max, LiveKit carries the audio, OpenAI hears you and works out the tutor's answers from the conversation, its board and what it knows of your studies, and ElevenLabs speaks them; what you say is checked as it is turned into text, like a typed message. On a teen account, Microsoft Azure AI Speech hears you and speaks the tutor's answers, and Anthropic writes them, each checked before it is spoken. In a typed conversation, Anthropic answers. Only the providers in the Processor & Service-provider List receive data, each for its stated role. None of these AI providers (Anthropic, OpenAI, Microsoft Azure AI Speech and ElevenLabs) receives your data until you allow it on the permission sheet, and you can turn each one off in Settings, under Privacy and legal (Privacy Policy, section 4).
- Our terms with Anthropic and OpenAI do not allow them to train their models on learner content. ElevenLabs, which speaks the cards you ask to hear in the Scholarch voice, the tutor's answers read aloud or spoken in a live conversation, and its replies in a guided voice conversation, is set on our account not to use the text it receives to train or improve its models, and it receives that text only after you allow it.
- The words of tutor conversations are kept for 30 days. Safety flags (section 6) are kept for 730 days. Notes the tutor saves for you, and your learning record, stay in your account until you delete them or your account is erased. The Privacy Policy gives every retention period.
- A lasting restriction of an account is reviewed by a person.
These safeguards reduce risk. They cannot remove every error or every harmful use.
8. Decisions and fair treatment
No decision that has a legal effect on you, or affects you as seriously, is made by AI alone. Scores and marks guide your practice; they are not grades or credentials.
Some decisions are made automatically by fixed rules, for example whether registration is open to you from your age and country. You can ask a person to review any automatic decision that refuses or limits your account, at appeals@scholarch.ai. In the United Arab Emirates, Federal Decree-Law No. 45 of 2021 Regarding the Protection of Personal Data (the Personal Data Protection Law) gives you the right to object to a decision made only by automated processing that has legal consequences for you or seriously affects you.
The checks look at what a message asks for, not at who you are. Restrictions take account of severity, context, repetition and risk. A lasting restriction is reviewed by a person and may be appealed through appeals@scholarch.ai.
9. Reporting
Use the feedback control in the product or report@scholarch.ai to report unsafe, wrong or inappropriate tutor behaviour. Use security@scholarch.ai for a vulnerability or an account-security concern. Reports are handled under the Privacy Policy.
10. Law and guidance we follow
Consumer law in every market where we offer Scholarch Education forbids describing a service in a false or misleading way: in the United Arab Emirates, Federal Law No. 15 of 2020 on Consumer Protection, as amended by Federal Decree-Law No. 5 of 2023; and in the United States, section 5 of the Federal Trade Commission Act (15 U.S.C. 45). This notice therefore describes only what the tutor does today, and marks what is coming. Some US state laws require a chatbot to make clear that it is not a person in the situations they cover, for example California's Business and Professions Code section 17941 and its companion chatbot law (section 22602). Scholarch always says that it is an AI. California's companion chatbot law also requires an operator it covers to keep a protocol for a user who expresses thoughts of suicide or self-harm, and to publish its details. Ours, for every account, is published in section 4, «Our crisis protocol»: the tutor answers with care, never gives methods, and refers the learner to 988 in the United States and, elsewhere, to the support and emergency numbers listed there; a message blocked as a request to cause harm, or one that cannot be checked, still gets the crisis line. New York's General Business Law, Article 47, sets a similar duty for the AI companions it covers; the same protocol, and the line every tutor surface shows, apply. California's law also requires an operator it covers to say that companion chatbots may not be suitable for some minor users (section 22604). We say it in section 6 and on the line every tutor surface shows (section 1).
We also build Scholarch with two guidance documents in mind: the UAE Charter for the Development and Use of Artificial Intelligence (2024) and the GCC Guiding Manual on the Ethics of Artificial Intelligence. Their themes include safety, fairness, privacy, transparency, human oversight and accountability. They are guidance, not certification, and they do not replace binding law, the Terms of Service or your mandatory rights.
11. Your part
Follow the Acceptable Use Policy. Do not put secrets, passwords or another person's personal data into a message or a card. Check important guidance before you act on it.
12. Review and contact
We update this notice before the tutor's role, the safeguards or the providers change in a material way, and when the applicable law changes.
Reports: report@scholarch.ai
Appeals: appeals@scholarch.ai
Security: security@scholarch.ai
Scholarch Limited · Masdar City Free Zone limited liability company · MC 14829 · Licensing authority: Masdar City Free Zone · Smart Station, First Floor, Incubator Building, Masdar City, Abu Dhabi, United Arab Emirates
Show the source hash
09be3d28d476b71579e62e718259730628360bc5e7d143a0678c2ce95ff08c7b