AI safety at IQChess
Version: 2026-10-27 • Last updated: 27 October 2026
IQChess uses AI to coach chess. This page explains what our AI features do, how we keep them safe for the young people who use them, and what schools and families can control. It forms part of our Privacy Policy, Terms of Service and, for schools and clubs, the Data Processing Agreement.
1) Our AI features
- Kai — a chess sparring partner the user plays against and can chat with during the game.
- Deep analysis — questions and follow-up questions about a chess position.
- Fast analysis, game reports and hints — one-shot explanations of a move or a game.
- Chess Gym grading and Think aloud — feedback on a chess analysis the user writes.
All of them are about chess. Chess facts in the answers (legal moves, threats, evaluations) come from a chess engine and code checks, not from the AI's guesses.
2) It is an AI, and it says so
- Kai's first message in every game says that Kai is an AI, not a person, and can make mistakes. If asked, Kai says plainly that it is an AI and never claims to have feelings or a life outside chess.
- During a long chat, Kai reminds the user at least every three hours that it is an AI and suggests a break.
- AI answers can be wrong. The app says so next to AI analysis, and our Terms of Service explain that AI output is not advice.
- AI chat, like any chatbot, may not be suitable for some minors. Schools and clubs can switch it off for their students (§6).
3) Self-harm and suicide protocol
Whenever a user types to one of our AI features — Kai, Deep analysis, Chess Gym or Think aloud — the message is checked before it reaches any AI model for words that express wanting to die or to hurt oneself, in every language the app supports.
- If the check matches, the message is not sent to the AI. The user instead sees a fixed message, written by people and shown in their language, that encourages them to talk to a parent, teacher, school counsellor or another adult they trust right now, to call their local emergency number if they may be in danger, and to find a free, confidential helpline in their country at findahelpline.com. It also reminds them that they are talking to an AI.
- The AI itself is instructed never to discuss suicide, self-harm or other harmful or adult topics, and to point a user who seems upset or unsafe to a trusted adult.
- We record only how many times this message is shown, and on which feature — never the user's words or identity — so we can monitor and report on the protocol as the law requires.
- The check deliberately ignores everyday chess language ("this move kills the attack", "suicide chess"). It can still occasionally show the message by mistake; we prefer that to missing a real case.
4) Keeping answers on chess
- Questions pass a safety check before an AI answers; requests outside chess, attempts to change the AI's instructions, and harmful content are refused.
- The AI is instructed never to ask for personal details (full name, school, address, age or photos) and not to send users to other websites or apps; links and web addresses are removed from Kai's replies.
- Answers pass an output check that removes internal details, links and claims about the board that the engine shows to be false.
5) Your data and AI training
- The AI providers we use (listed in the Privacy Policy §6a) receive the chess position and the typed question — never the user's account identity — and may not use it to train their models.
- Students enrolled by a school or club: nothing they type, and no AI answer to them, is ever used to train or improve AI models — ours or anyone else's.
- Other users: de-identified conversations (with no account information attached) may be used to evaluate and improve our own chess-coaching models, as described in the Privacy Policy §4.
6) What schools and clubs control
In the teacher dashboard (Settings → Safety and Student app), an administrator decides for every student of the organisation:
- Kai chat — starts switched off for schools and in young-learners mode (ages 6–9).
- Deep analysis — starts switched off in young-learners mode.
- AI grading of written answers (Chess Gym, Think aloud).
Each switch is enforced on our servers, not only in the app. Kid-safe mode, separately, limits friends and games to classmates and hides last names and emails. Teachers do not see students' AI chats.
7) Contact
Questions or concerns about AI safety: privacy@iqchess.org — 9517-9206 Québec inc., 4736 Saint Jean Blvd., Pierrefonds, QC H9H 4B2, Canada. If you are worried about someone's safety right now, contact local emergency services.