How Classroom Board uses AI
Classroom Board is a screen a teacher shows to a class. It has four AI helpers, all for the teacher: Fill with AI, AI Chat, AI picture and Agent Daisy. Students do not have accounts and do not type into them. This page says plainly what each one does, what is sent, what checks run, where a person stays in charge, and what the limits are. Agent Daisy, also for the teacher only, plans a lesson from a brief the teacher types (up to 300 characters) and the year level, learning area, lesson length and class needs the teacher picks, then makes the lesson's boards, using Fill with AI and AI pictures for the parts that need them. Classroom Board does not screen Daisy's brief, its plan, its fills or its picture descriptions for words, names or web addresses. The AI provider's own safety rules still apply, and Daisy's pictures keep the nudity screen and the real-person reminder. You check Daisy's draft before it is built and the lesson before you show it. The Live Arcade (class games) adds one more use, also for the teacher only: AI can write a set of quiz questions from a topic the teacher types. Students who join a game answer those questions; they never use AI.
The four AI helpers
| Helper | What it does | What is sent to the AI provider |
|---|---|---|
| Fill with AI | Suggests the contents of a widget: a word bank, a set of quiz questions, the cells of a graphic organiser, or the Daily Quiz, Daily Starter, Daily Review and Activities items on your own topic. Nothing lands on your board until you press Use this, except in a lesson Agent Daisy builds, which lands as a new lesson for you to read. | The topic you type (up to 120 characters), the year level, and the options you chose (for example word level or question type), inside our fixed instruction. Also, for a Daily Review, up to five focus points you type; if you chose a curriculum code, its description from the published curriculum; and for a lesson Agent Daisy builds, Daisy's notes on that lesson. |
| AI Chat | A chat you run in front of the class. You type; the AI replies; you can ask it to critique its own last reply. Replies are hidden until you show them (you can turn that off). | The messages you typed in the conversation (each up to 1,000 characters), earlier replies in the same conversation, and the year level, inside our fixed instruction. |
| AI picture | Makes a picture from your description, or the topic of a See, think, wonder card, when you press Show a picture, and saves it to My pictures. It can be used on a board or as a background. A card's picture goes onto the board beside the card: one press, one picture. | Your description (up to 300 characters), or the card's topic (cut to 300 characters), and the style and shape you chose (for a card, a wide photo), inside our fixed instruction. Before any picture is made, the same description is also sent to our text AI provider to check whether it asks for a real, identifiable person, so a reminder can be shown with the picture. Pictures of real people are allowed. |
| Agent Daisy | Plans a whole lesson from your brief, then fills its parts and makes its pictures for you to check. | The brief you type (up to 300 characters), and the year, subject, lesson length and class needs you chose, inside our fixed instruction. The brief is not checked for class-list names, blocked words or web addresses. |
Not sent: your name, your email, your teacher ID, your class lists, your boards or lessons, or any student record. We send only the requests described on this page. Agent Daisy's notes on a lesson go with that lesson's fills.
Student names
- Class lists can hold student names. We ask you to use first names or initials, or to leave names out.
- Before an AI request leaves your device, Classroom Board checks it against the first names on your class lists and in your Catch-up widgets, except when Agent Daisy plans and builds a lesson. The same check runs again on our server, including on earlier turns of a chat, and both also know the names typed in your Name picker quick lists. If a name is found, the request is refused and nothing is sent to the AI provider. Agent Daisy's planning and building are not checked this way: the brief, Daisy's plan and what Daisy makes for the lesson are sent as written, so a student's name typed there goes to the AI provider. A first name followed by another capitalised word that is not on your lists is treated as a public person's name (for example a famous athlete's full name) and is not refused, so please do not type a student's full name into an AI box. A chat reply is also checked for those names before it appears, and is held back if it has one. The AI is not asked to remember or look up names.
- This is a safety check, not a promise. It matches names that are on your lists. If you type a student's name that is not on a list, the check will not know it is a student. Please do not put student names into any AI helper.
- We do not log what you typed, the AI's reply, a name, or your teacher ID. Our logs record the kind of request, the result (for example allowed, refused, over the daily limit), and numbers such as timing.
Four checks that reduce the chance of unsuitable output
These checks help. They do not make the output safe, and they do not replace your judgement.
- Word screens. A list of plain unsuitable words (sexual, violent, self-harm, drug, swearing and slur terms) is checked against what you type. For Foundation to Year 2 the list is longer. The same base list is checked against what comes back. Agent Daisy's brief, plan and fills are not screened this way.
- A classroom instruction. Every request carries an instruction written for an Australian classroom: Australian English, plain words for the year level, nothing frightening, violent, sexual, or about self-harm, drugs or weapons, no personal details about real people. Agent Daisy's plan carries its own planning instruction instead, which names people where the lesson needs them, teaches a hard topic plainly for the year level, and never gives a real person's private details. For AI Chat the instruction also says the class is watching, and to say when it is not sure.
- The AI provider's own safety rules. The providers refuse some requests themselves. If they do, you see "I can't help with that one in a classroom" or a similar plain line.
- A check of text answers. What the AI writes back (a Fill with AI result or a chat reply) is screened for the same blocked words before it appears; what it writes for an Agent Daisy lesson is not. A chat reply is also checked for class-list names.
For AI pictures, checks 1, 2 and 3 apply (the screen on your description, with a short extra list of nudity words for pictures; the classroom instruction; and the provider's own rules), plus a check of whether the description asks for a real, identifiable person, living or from history. A picture in an Agent Daisy lesson skips the word list but keeps the nudity words and the real-person check. Pictures of real people are allowed: when the check finds one, or cannot answer, a small Safety reminder shows beside the picture (see below). We do not scan the finished picture ourselves. You see it as soon as it is made and can remove it from My pictures.
Pictures of real people
A teacher can make a picture of a real person, for example a made-up scene from history. When our check finds a real person in the description, or cannot tell, a small Safety button shows beside the picture, never on it. Pressing it shows this reminder:
About pictures of real people
This is a made-up picture of a real person. It is not a record of what happened.
Made-up pictures can carry bias and stereotypes. Ask the class what the picture might get wrong.
Treat real people with respect.
Compare it with real sources. You can add a real photo beside it with a Picture widget.
For any AI picture
AI pictures can get details wrong, such as faces, hands, words and places.
Tell the class it was made by AI.
Look at it yourself before the class does.
The button is named "Safety reminder for this picture" for screen readers.
A person stays in the loop
- Fill with AI: a draft appears for you to read. It only goes onto the board when you press Use this, and you can edit it; for Daily Starter, the cards can be edited on the card, AI-made and ready-made alike, or you can fill again or press Back to ready-made. A lesson Agent Daisy builds is different: you read Daisy's plan first, then the filled lesson lands as a new lesson, not on a board you have open. Read it before you show it.
- Daily widgets on your topic: in the Daily Quiz, Daily Starter, Daily Review and Activities, items you make with Fill with AI carry no label on the board, the slides or the print sheets; in Edit, a routine of your own is marked Our topic. We ask you here and on the Help page instead, in the words on the Help page: "Items you make with Fill with AI on your topic in the Daily Quiz, Daily Starter, Daily Review and Activities are made by AI, so they can be wrong. Check them together with the class."
- AI Chat: replies are hidden until you choose to show them (the default). Only you type. The widget has no on-screen label about checking; we ask you here and on the Help page instead, in the words on the Help page: "AI replies can be wrong. Check them together." A Critique this button asks the AI to point out what might be wrong or missing in its own reply, so the class can check it.
- AI picture: from your description, or the topic of a See, think, wonder card, when you press Show a picture. You see the result and choose whether to use it. A card's picture goes onto the board beside the card as soon as it is made, one press, one picture; Show a picture is in Edit only, never in Present, and you can remove the picture.
- Daily limits apply to every teacher (for Fill with AI, AI Chat messages and AI pictures) and per-minute limits slow rapid-fire requests.
- You are responsible for what is shown to your students. Please look at AI results before the class does.
Which AI models, and the rule for AI Chat
Fill with AI, the real-person check on pictures and the Live Arcade's question sets use a small, fast model from Anthropic. AI Chat uses a small, fast model from each provider, Anthropic and OpenAI, and you choose which. Agent Daisy plans a lesson with a larger Anthropic model, then fills its parts with the same small model as Fill with AI. Pictures are made with an OpenAI image model. We do not name versions here because they change. AI Chat is for support and modelling in class, where the stakes are low, and we keep it on small, fast models. We do not move AI Chat to larger, more capable models.
We name the providers because a school should know who processes the text:
| Provider | What it does for us | Where requests go |
|---|---|---|
| Anthropic | Writes text for Fill with AI, Agent Daisy, the Live Arcade's question sets and AI Chat when you choose Claude; checks each picture description for a real person | Anthropic's default service, which processes requests globally; no regional option is set. Anthropic is a United States company. |
| OpenAI | Writes AI Chat replies when you choose OpenAI; makes AI pictures | OpenAI's United States service (api.openai.com). OpenAI is a United States company. |
What the providers say about your content
We read each provider's published terms on 10 November 2026. Please check them yourself too, because they change.
- Not used to train their models by default. OpenAI: "data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in)" (developers.openai.com/api/docs/guides/your-data). Anthropic: by default it does not use inputs or outputs from its commercial products, including its API, to train its models (privacy.claude.com, "Is my data used for model training?"). We have not opted in to either.
- Short-term retention for safety. OpenAI keeps abuse-monitoring logs of API requests for up to 30 days by default. Anthropic's retention depends on the model and is set out in its own terms. Each provider may keep requests for a limited time under its own terms, for example to check for misuse.
- We therefore do not say that providers keep nothing, and we do not claim zero retention. We say they do not train on your content, and may keep requests for a limited time under their own terms.
Alignment with the Australian Framework for Generative AI in Schools
The Framework (Education Ministers, 2023; 2024 review endorsed June 2025) has six principles and 25 guiding statements. It is for schools, not vendors. We have read it as the measure a school will hold us against. Here is how we match it, honestly:
| Principle | What we do | Where we are not there yet |
|---|---|---|
| 1. Teaching and Learning (statements 1.1 to 1.6) | The AI is optional, teacher-run and visible to the class. AI Chat's "Critique this" and hidden replies are built to model careful checking (1.3, 1.4). Fill with AI lands only when the teacher presses Use this; a lesson Agent Daisy builds lands as a new lesson for the teacher to read first. | We do not teach students about AI or its biases (1.2); that is the school's job. The AI helpers do not design assessments (1.5) or manage academic integrity (1.6). |
| 2. Human and Social Wellbeing (2.1 to 2.3) | The four checks and the classroom instruction aim to avoid harm (2.1). The teacher decides what is shown. No AI companion; no student-facing AI. | We have not run a bias test of Classroom Board's AI output (2.2). Our word screens are short and plain on purpose; they do not catch everything. |
| 3. Transparency (3.1 to 3.3) | This statement and the Help page ("AI replies can be wrong. Check them together."); AI Chat has no on-screen label. Daily items made with Fill with AI on your topic have no on-screen label either; Help says they are made by AI and to check them together. We name the providers and explain what is sent (3.1, 3.3). | Telling families how AI is used in class is the school's disclosure under 3.2; we do not contact families. This statement is written so a school can share it with them. |
| 4. Fairness (4.1 to 4.4) | We are working towards WCAG 2.2 AA (see our Accessibility Statement) (4.1). Australian English throughout (4.3). | Regional and remote use (4.2): the board needs the internet for AI and for first sign-in; it keeps working for a time offline. Cultural and Indigenous intellectual property (4.4): not yet reviewed in AI output. |
| 5. Accountability (5.1 to 5.4) | Teachers stay in control of decisions (5.1). The checks are covered by automated tests that run before each change to the code is accepted (5.2). You can tell us about an unsuitable result by email (5.4). | We do not yet have an in-product "report this result" button or a published incident register (5.3, 5.4). Testing so far is functional and real-server, not a formal AI risk assessment. |
| 6. Privacy, Security and Safety (6.1 to 6.5) | Student names are optional, stay in your account, and are checked out of AI requests, except when Agent Daisy plans and builds a lesson (6.1, 6.3). We do not log prompts or replies. Our Privacy Policy and this statement say what is collected and shared (6.2). Sessions and access are controlled (6.4). | We do not yet delete Classroom Board data automatically when an account closes; the teacher uses Delete my data or emails us (6.1). Copyright for what teachers upload is the teacher's responsibility under our Terms (6.5). We have not had an independent security test. |
eSafety
The eSafety Commissioner's age-restricted material codes for designated internet services, including generative AI services, came into effect on 9 March 2026. We have not concluded whether or how they apply to a teacher-operated classroom tool. The measures above (a classroom instruction, word screens, the provider's refusals, a text check, a teacher in control, no student accounts) are the kind of controls those codes expect of AI services. AI Chat is not a companion and is not a chat for children.
Limits
- AI can be wrong. It can be inaccurate, out of date, one-sided or unsuitable. Check what it says before you rely on it or show it.
- Word screens are short lists and will miss things. Providers' refusals are not perfect.
- The name check only knows first names on your class lists, Catch-up lists and Name picker quick lists, and it lets a first name through when it is followed by a capitalised word that is not on your lists. It does not know a name you have only typed somewhere else, such as a text widget. It does not run when Agent Daisy plans and builds a lesson.
- We cannot control what a provider's model produces.
- AI Chat has no memory of your class beyond the conversation on screen.
Report a result
If an AI result is unsuitable, email classroomboard.au@gmail.com with "AI Output Concern" in the subject and: which widget, roughly when, and what you asked (leave out names). We keep a record and use it to improve the checks. Please do not send student information. We look at a report of serious AI output within 1 business day. If a confirmed AI incident affects a school that has given us a contact, we tell that school within 2 business days of confirming it.
AI question sets for class games (Live Arcade)
In the Live Arcade, the teacher runs a class game on the board, and students join it on their own devices with a PIN from the board.
What it does. A teacher types a topic and a year level and asks for a set of quiz questions. The AI writes a draft set. It opens in the editor, where the teacher reads it, changes it and decides whether to save it. Nothing is shown to a class until the teacher starts a game with it.
What is sent. The topic and year level the teacher typed, and how many questions and which kinds the teacher chose. Never anything a student did in a game, and never a class list; the topic is checked for names, as below.
The checks. The same ones every AI fill gets: the topic is refused if it holds a first name from one of the teacher's class lists or a blocked word; the answer is checked against a fixed shape and screened for blocked words before the teacher sees it; each set counts as one of the day's AI fills.
Students. Students never use AI in a class game. They answer the questions their teacher chose.
Monitoring, working with schools, and switching AI off
Schools run their own monitoring, and we give them a structure for it. A school can name an AI contact by emailing classroomboard.au@gmail.com with "School AI contact" in the subject; we use that contact for notices of AI incidents and AI changes. Each term, a school can ask the teachers who use Classroom Board three questions: has an AI result been wrong or unsuitable, has a student's name gone into an AI request, and has anything AI-made been shown to a class without being read first? Tell us what you find. We acknowledge a school's concern within 2 business days and look at a serious one within 1 business day. On request, once a term, we hold a short check-in with the school's AI contact. We tell the school's AI contact about a significant AI change, in the release notes and by email, at or before release. AI runs only when you press for it, and the board works without it. There is no switch today that turns AI off for a teacher or a school. A school that does not want its teachers to use AI can ask them not to use the AI helpers.