DPIA template: using Monty's classes
10 November 2026. A data protection impact assessment a school can adapt, following the ICO's DPIA template's steps, filled in with what Monty knows about its own processing. The school, as controller, owns the DPIA: its DPO advises, and someone senior signs it off. Text in square brackets is for the school. Monty's facts come from the record of processing and the agreement; our reading, not legal advice.
Step 1: why a DPIA
Monty processes children's data, uses an AI, and sends some data to the United States. The ICO lists children's data and innovative technology among the reasons to carry out a DPIA, and KCSIE 2026 asks schools to check AI products against the DfE's product safety standards before using them.
Step 2: the processing
- What: pupils aged 11 to 18 learn Python in a browser. Joining a teacher's class lets the teacher see their progress, and lets Monty send a safeguarding concern to the school's DSL.
- Pupils' data on Monty's server: a generated username and a chosen nickname (never their name, email or date of birth); class membership; progress; minutes active a day; counts of large pastes, worked examples and AI calls (with how many were chat and late at night); a keyed hash of their login code; a keyed fingerprint of a linked school Microsoft or Google account, if they link one (built, and not switched on today); a flagged message, if one is raised.
- Staff data: work email, roles, an encrypted authenticator secret, a keyed fingerprint of a linked account; the DSL's and deputy's email and optional name; an access log (ids only).
- Not on Monty's server: pupils' names (they stay in the teacher's browser); chats (only a flagged message is kept).
- Who sees what: teachers see their own classes' progress and counts, never chats or flags; the DSL sees flags and a weekly count of patterns; admins see every class for cover; Monty's operator sees what's needed to run the service, and a flagged message only by choosing to, which is logged.
- Where: Vercel (hosting) and Neon (database) in the United States; Anthropic (the AI) in the United States, sent the exercise, the pupil's code and messages but no username, nickname, class or school; Amazon SES (email) in London.
- How long: classes until the end of the first month of the school's next year (30 September where the year starts in September); alerts 30 days after the DSL confirms them; the access log 24 months; see the agreement's clause 9. The periods are the same for every school.
- Lawful basis: [public task, Article 6(1)(e), for a maintained school or academy; legitimate interests, 6(1)(f), for an independent school: the DPO decides]. For a flagged message that reveals special category data: [substantial public interest, safeguarding of children, Schedule 1 paragraph 18 of the DPA 2018: the DPO decides].
Step 3: consultation
[Who the school consulted: the DPO, the DSL, IT, the head of computing, pupils or parents, governors.] Monty's operator answers questions; the DfE's standards page and the trust page set out the rest.
Step 4: necessity and proportionality
- Pupils are known by username and nickname, so a breach of Monty's server names nobody.
- Only progress and counts reach teachers; chats are never stored, and teachers never see flags.
- The AI is sent no identifying data, and never trains on it (Anthropic's terms).
- Pupils are told what their teacher sees when they join and again each half term.
- Teachers can delete a pupil or a class at once; classes delete themselves after the year.
Step 5: risks
| Risk | Likelihood | Severity | Monty's measures |
|---|---|---|---|
| A pupil is identified from their username, nickname or progress | [Remote] | [Minimal] | No names on the server; nickname rules warn against real names |
| Someone else signs in as a pupil | [Possible] | [Minimal] | Cards with a code; lockouts after wrong codes; "sign out everywhere"; shared-computer mode |
| A member of staff's account is taken over | [Remote] | [Significant] | An authenticator code at every sign-in; sessions ended by removal; access logged |
| A safeguarding concern is missed or delayed | [Possible] | [Severe] | The AI and crisis phrases; an alert within 5 minutes; retries and delivery checks; weekly patterns |
| Data sent to the United States is accessed by others | [Remote] | [Significant] | The EU clauses with the UK addendum; no identifying data sent to the AI |
| A pupil's message is kept longer than Monty keeps it | [Possible] | [Minimal] | Anthropic keeps a message its own safety systems flag for up to 2 years; nothing identifying is sent |
| The AI writes something harmful or wrong | [Possible] | [Significant] | Safety instructions, the eval before changes, Report on every reply, tests decide |
| Pupils rely on the AI instead of thinking | [Possible] | [Minimal] | No answers, hints first, offloading signals, the AI level per class |
Step 6: measures to reduce risk
[The school's own: for example, a lesson on what Monty is before first use; the DSL added at setup; the AI set to "hints only" for assessments; the weekly patterns email kept on.]
Step 7: sign-off
| Name and date | Notes | |
|---|---|---|
| Measures approved by | ||
| Residual risks approved by | ||
| DPO advice | ||
| DPO advice accepted or overruled by | If overruled, why | |
| Review | At least yearly, and when Monty tells the school of a change |