Last updated: 28 September 2026 · Next review: by September 2027 · Contact: hello@emblemforall.co.uk
This statement sets out how Emblem measures up to the Department for Education's Generative AI: product safety standards (updated 19 January 2026), standard by standard, including where it does not yet meet one. Emblem meets most standards through its design: pupils never chat with an AI, and every AI-written sheet is checked by a teacher before a pupil sees it. This page is Emblem's published AI safety policy.
Emblem is a teacher tool for UK primary schools (ages 4 to 11). It adapts a teacher's lesson so pupils with additional needs, including SEND and EAL, can work on the same learning question as their class. It then delivers the sheets on paper or tablet and records the support given and its impact. Pupils are identified by a symbol, never by name.
DfE use cases: 1 (content creation and delivery) and 2 (personalised learning and accessibility). The support-plan summaries serve use case 7, but are produced by the app's own calculations, not by AI.
| Feature | Who uses it | Generative AI? |
|---|---|---|
| Adapting a lesson into sheets, and editing a sheet on request | Teachers | Yes (Claude, by Anthropic) |
| Teacher answer keys and TA pre-teach slips | Teachers | Yes |
| Matching a specialist's advice to supports | SENCO or teacher | Yes |
| Completing a sheet on a tablet or laptop | Pupils | No. The sheet was written in advance and checked by the teacher. |
| Tap a word for a child-friendly meaning | Pupils | Yes, limited to a single word. The only AI feature pupils use. |
| Marking, outcomes, fading suggestions, summaries | Teachers | No |
The learner-facing standards (3, 4, 10, 11 and 12) therefore apply to one narrow feature, the word helper, and to the design of the sheets pupils receive.
Emblem meets 6 of the 13 standards, mostly meets 4, and partly meets 3. None is unaddressed.
| # | Standard | Status | In one line |
|---|---|---|---|
| 1 | Stated purpose | Meets | Purpose, ages and SEND focus are stated; no impact claims beyond the evidence |
| 2 | Educational use cases | Meets | Use cases 1 and 2 |
| 3 | Filtering | Meets | Pupils can send only a single word to the AI; there is no free text or chat |
| 4 | Monitoring and reporting | Partly meets | Staff see every word lookup, refused ones first; no DSL alerts yet |
| 5 | Security | Partly meets | Strong pupil isolation; staff share one school code, no MFA yet |
| 6 | Privacy and data protection | Mostly meets | No names, DPIA, no model training; pupil-facing notice is on paper, not in the app |
| 7 | Intellectual property | Meets | Pupil work never goes to the AI; content is never used for training |
| 8 | Design and testing | Partly meets | Built with teachers and tested in pilots; no formal testing with pupils yet |
| 9 | Governance | Mostly meets | Named lead, complaints route and risk assessment; this page is the AI safety policy |
| 10 | Cognitive development | Mostly meets | Scaffolds, never answers, and supports fade; no published impact plan yet |
| 11 | Emotional and social development | Meets | No chatbot, persona or conversation |
| 12 | Mental health | Mostly meets | No channel for pupils to disclose to AI; protocol below; no clinical adviser |
| 13 | Manipulation | Meets | No ads, upsells, flattery or engagement tricks |
Emblem is for teachers and SENCOs in UK primary schools, adapting lessons for pupils aged 4 to 11 with SEND or EAL, across all subjects. Its design draws on EEF guidance on SEND in mainstream schools and DfE policy. Emblem has not been independently evaluated, and makes no claim about its impact on attainment. It is in pilot use in three schools. This statement is reviewed whenever a feature changes what Emblem is for.
Use case 1 (content creation and delivery) and use case 2 (personalised learning and accessibility). See the table above for which features use AI.
Harm is prevented by design rather than by filtering afterwards. Pupils cannot type messages to an AI. The one pupil feature that uses AI, the word helper, accepts only a single word of letters (up to 30 characters) and returns a two-sentence, child-friendly meaning from a fixed prompt. Words unsuitable for a primary child get the reply "Ask your teacher about this word." There is no conversation, so the filtering does not weaken over a long exchange. Meanings are cached and shared, so the same safe answer is reused. Every sheet a pupil sees was written in advance and checked by their teacher, who can edit or withdraw it.
Because pupils have no free-text channel to the AI, a safeguarding disclosure cannot be made to the AI. Everything pupils write goes to their teacher, who marks it. Pupils' check-in ratings and "how did that feel?" answers are visible to staff. Every word a pupil sends to the word helper is logged with their symbol, class and time for 90 days. Staff see these in the Work tab, with refused words and anything that was not a single word listed first. Not yet in place: automatic alerts to a named DSL, a DSL contact set at set-up, and usage-duration reports. A DSL contact with alerts is next on the roadmap. Cognitive-offloading reporting does not apply: pupils have no way to ask the AI for answers.
All traffic is encrypted and all secrets are held on the server. If the service is ever misconfigured, it refuses every request rather than letting anyone in. Each pupil device has its own token that opens only that child's work, enforced by the server; pupil devices cannot generate content or reach staff screens. Wrong laptop codes are rate-limited and AI use has a daily cap per school. Jailbreaking the word helper is not possible through its single-word input. Gaps: staff share one code per school rather than individual accounts, so there is no multi-factor sign-in or per-user permission level. Leadership views are separated by a PIN within the staff app. Emblem does not yet hold Cyber Essentials. Individual staff accounts and Cyber Essentials are planned. New versions are tested before release (see standard 8).
No pupil names are stored. The school is the controller and Emblem the processor, under Article 28 terms. A DPIA template is provided, pre-completed. Data is stored in London. Lesson content only is sent to Anthropic in the US, under the UK International Data Transfer Addendum, and deleted within 30 days. No pupil or school data is used to train or fine-tune models, or for any commercial purpose. Child-friendly and parent privacy notices are provided for schools to share. Gap: the notice is not yet shown to pupils within the app itself. See our privacy policy; the data protection pack for schools is available on request.
Pupils' work is never sent to an AI model. Teachers' lesson content is sent only to generate that teacher's sheets, is not used to train models (Anthropic's commercial terms), and is not used for Emblem's own product development. The one exception is a problem report a teacher chooses to send, which holds the learning question and a short note, is used only to fix that fault, and is deleted after 90 days.
Emblem is designed by a serving primary deputy headteacher with SEND leadership experience, and refined through use in three pilot schools. Automatic checks run on every AI-written sheet and flag broken items (for example, a "circle the answer" question with nothing to tap) for the teacher. Each release is tested against realistic lessons and a mocked AI service before it goes live. Gap: there is no documented user testing with pupils yet, and testing is not yet independently reviewed.
Edd Chalkley, founder, is accountable for AI safety decisions. The risk assessment is the DPIA in the data protection pack. Complaints and safety escalation are set out below. This page is Emblem's published AI safety policy, and will be kept current.
Emblem's core rule is that every child works on the same learning question as their class. The support changes, never the demand. Pupils cannot ask the AI to answer their questions, and sheets are written to leave every item for the pupil to answer: a rule given to the AI, and checked by the teacher before release. Where a worked example is chosen as a support, it is a separate model problem followed by a "we do together" step, following EEF guidance on explicit instruction. Supports are designed to fade: when a pupil meets the learning question independently three times in a row with a support, Emblem suggests reducing it, and the teacher decides. The word helper explains a word, not an answer. Gaps: there is no published child-development impact plan or record of expert oversight yet. Cognitive-offloading tracking does not apply, since pupils have no way to ask for answers.
Emblem has no chatbot, character, avatar or name for its AI, and does not speak in the first person. Pupils never converse with it. The only question pupils are asked about their feelings is a fixed, optional choice about the sheet they have just done ("How did that feel?" with three faces). The answer goes to their teacher and is never processed by AI. Time limits are not built in, because pupils do not use AI in sessions: a word lookup is a single tap.
Pupils have no free-text route to the AI, so they cannot disclose distress to it or receive AI advice about it. What pupils write in their work and check-ins goes to their teacher, a human, who follows the school's safeguarding policy. The protocol below sets out what Emblem does if a concern reaches it. Gap: child mental health expertise has not yet been formally involved in the product's design.
No advertising, no upselling, no paid options shown to pupils, no dark patterns and no engagement tricks. Pupils see a simple count of learning questions they have met, as a transparent, low-stakes encouragement. Sheets contain plain task instructions for children, with no praise or persuasion written by the AI.
Emblem does not replace a school's safeguarding procedures. Pupils' words reach their teacher, not an AI, so the school's own policy applies to anything a pupil writes.
| What | How | Response |
|---|---|---|
| A safety issue with generated content | "Report a problem" on the sheet, or hello@emblemforall.co.uk with "Safety" in the subject | Reviewed the same working day; protocol step 4 if needed |
| A complaint or data protection question | hello@emblemforall.co.uk | Acknowledged within 2 working days, answered within 10 |
| Not satisfied with the answer | Escalate in writing to Edd Chalkley, founder | Final response within 20 working days |
| Data protection concerns at any stage | The Information Commissioner's Office |
This statement is reviewed at least once a year, whenever a feature changes what Emblem does with AI, and whenever the DfE updates the standards.
Planned improvements, in order: a DSL contact at set-up, with alerts when the word helper refuses a word (standard 4); individual staff accounts with optional multi-factor sign-in (standard 5); Cyber Essentials (standard 5); documented user testing with pupils (standard 8); a published child-development impact plan (standard 10); input from a child mental health professional (standard 12).