How the Safety Guard works

1. A class is recorded (audio only) when every student in it has consent.
2. After the class, the audio is turned into text, in whatever language was spoken.
3. An AI reads the text, the class chat and, for students, private messages. It looks for risky or off-topic parts.
4. Anything it finds is shown to the academy's admin as a flag, with the time, the exact words and a short reason in plain English.
5. The admin plays that part, then decides: dismiss, warn, or escalate. Every decision is logged.

• Asking for contact details, photos or a private call
• Asking a child to keep something secret
• Sexual or inappropriate talk
• Threats or abuse
• Hate speech
• Harsh language to a child
• Chatting that has nothing to do with the lesson

Recitation and teaching of the Quran are normal and are not flagged (when the academy teaches Quran).

• It never punishes, blocks, removes or contacts anyone by itself.
• It never decides alone. A person at the academy makes every decision.
• It does not look at video, because there is no video.
• It does not check classes of students without consent.
• It does not check public meetings, which are never recorded.

The AI is not perfect. It can miss something, and it can flag something harmless. That is why a person always checks. We test it on made-up classes before every change to its instructions.

The audio and text of a class are sent to OpenAI only to be turned into text and checked. Recordings, transcripts and chat are deleted after the academy's retention time (30 days by default). Only the academy's owner and admins can open them, and every access is logged. More in the privacy policy.