← Back to vespa.academy
AI Usage Policy
Last updated: August 2026
Summary: VESPA Academy uses AI as an optional, advisory tool to support student coaching and related learning features. Portal AI features run when a user actively requests them. The minimum necessary prompt data is sent via encrypted API to OpenAI (primary student/staff portal AI) and, for 4Sight’s own CRM/operations tools, Anthropic. We do not train AI models on school or student data. OpenAI organisation controls for our API workspace include disabled API call logging, disabled hosted tools, and disabled data sharing for model improvement (evidenced in our compliance pack, August 2026).
August 2026 update: This policy was corrected following an internal forensic audit so that published descriptions match the live product. Earlier versions understated the number of AI features and incorrectly stated that names and conversation history were never sent to AI APIs. The corrected position is set out below.
1. Overview
VESPA Academy (“the Platform”), operated by 4Sight Education Ltd (“we”, “us”), incorporates artificial intelligence features to enhance student coaching and related educational tools. This policy explains how AI is used, what data is involved, and the safeguards in place.
This policy should be read alongside our Privacy Policy, Student Privacy Policy, and Data Processing Agreement.
2. AI Features in VESPA (student / school portal)
The following AI features are available in the school-facing Platform. All are optional and initiated by a user action (button or chat send). They do not run on login or as silent background jobs for students.
| Feature |
Purpose |
Who uses it |
What is typically sent to the AI API |
| Coaching Narratives |
Draft coaching summaries to help tutors write personalised feedback |
Staff |
VESPA scores, selected questionnaire items, free-text reflections, year group, and (where available) the student’s first name so the draft can be personalised. Staff still review before sharing. |
| UCAS Statement Feedback |
Structured feedback on personal statement drafts |
Students |
Draft statement text (and related in-app tutor comments where relevant). Email/name are not required in the model prompt. |
| UniGuide AI Advisor |
Multi-turn conversation to explore university / course options |
Students |
First name (for greeting/context), school name, year group, subjects, preferences, the latest message, and recent conversation history for that session (loaded from our database and passed into the model). History is stored against the student account in VESPA. |
| Study Planner AI |
Suggests a weekly study plan from the student’s prompt |
Students |
Week start date, the student’s prompt text, and a display name where available (or a generic label such as “Student”). Email addresses are not sent to the model. |
| Study session assist |
Short session coaching ideas for a single study block |
Students |
Subject, topic, exam board / level, and study-idea context — not email. |
| Coaching Flow Assist |
Helps staff prepare coaching conversation prompts |
Staff |
Coaching context such as reflections/goals/scores as provided for that flow. |
| Deck Builder AI |
Helps generate slide / activity deck content |
Staff |
Topic and generation instructions for the deck. |
| Homework / activity AI |
Generate, mark, or feedback support for homework and portfolio tasks |
Staff / students (feature-dependent) |
Task content, answers, and marking context needed for that request. |
| FL4SH card generation |
Generates revision flashcards where enabled |
Students / staff (feature-dependent) |
Card-generation prompts; identity for the feature is stored in VESPA where the lite integration is used. |
2.1 CRM and internal AI (4Sight only)
4Sight also uses AI (primarily Anthropic Claude, with OpenAI as fallback in some tools) inside our own CRM and finance operations — for example school website enrichment, drafting commercial emails, and internal finance assistance. That processing is controller-side for 4Sight (our business operations). It is not student portal coaching and is not part of the school’s day-to-day student data processing in the Platform UI.
3. How AI Is Used
- Advisory / decision-support only. AI outputs are suggestions, summaries, or feedback. They must not be the sole basis for high-stakes decisions about a student without human review. No AI feature automatically changes grades, creates disciplinary referrals, or withdraws access to services.
- Human-initiated (portal). Portal AI features run when a user clicks or sends a message. They do not run silently on student login.
- Human-reviewed. Staff coaching outputs are for tutor review. Student-facing feedback is presented as guidance for the learner to consider.
- Conversation memory (UniGuide). UniGuide is intentionally multi-turn. We store the conversation in our database and re-send recent turns to the model so advice stays coherent. That is a product feature, not OpenAI-side conversation state.
4. AI providers and sub-processors
We do not train, fine-tune, or host foundation models ourselves. We call commercial APIs.
| Provider | Models (typical) | Role |
|---|
| OpenAI | GPT-4o, GPT-4o-mini (and equivalents) | Primary provider for portal AI features listed above |
| Anthropic | Claude Sonnet (API) | Primary provider for 4Sight CRM / internal AI; may be used as an alternative where configured |
- API processing. Prompt data is transmitted over encrypted HTTPS (TLS) so the provider can return a completion. We do not sell school or student data for advertising or unrelated commercial use.
- Sub-processor relationship. OpenAI and Anthropic are listed as sub-processors in our Data Processing Agreement (Schedule 2).
- International transfers. API processing may occur outside the UK/EEA (commonly the United States). Transfers are protected by Standard Contractual Clauses and the providers’ Data Processing Agreements.
5. Data and training — what we do and do not do
No school or student data from VESPA is used by 4Sight to train any AI model. Prompt data is sent to our AI sub-processors for processing the request. That is different from saying data never leaves our systems.
- We do not train models on school or student data.
- OpenAI API / organisation controls. Our OpenAI organisation is configured so that API call logging is disabled, hosted tools (web search, code interpreter, etc.) are disabled, and sharing of inputs/outputs and feedback for model improvement is disabled. Audit logging of admin/configuration changes is enabled. Under OpenAI API business terms, API data is not used to train their models.
- Anthropic API. CRM/internal features call the Messages API with Claude Sonnet models. We do not use Anthropic Files, Batches, code execution, or MCP connectors in our product code paths.
- No reuse by us for unrelated purposes. We do not sell prompts or outputs as training datasets.
6. Data minimisation (accurate position)
- We send the data needed for the specific feature to work well — not a full student MIS extract.
- We do not send demographic fields such as ethnicity, SEN, FSM, or disability status to AI APIs in the features reviewed for this policy.
- We do not send student email addresses to OpenAI for Study Planner generation (remediated August 2026). Email may still be used inside our own systems for authentication and to store UniGuide history against the correct account.
- Some features do use a student’s first name (and, for UniGuide, school / year / subjects) so the experience can be personal and relevant. Where a name is not available, a generic label is used.
- Model parameters (including temperature) vary by feature to balance usefulness and consistency; they are not fixed at a single value for every call.
7. Bias and Fairness
- System prompts are written to be inclusive, encouraging, and constructive.
- UCAS feedback is structured around statement content, not protected characteristics.
- Providers include safety layers designed to reduce harmful outputs.
- We review prompts and sample outputs when we ship or change AI features.
8. Safety and Content Controls
- AI-generated content is presented as suggestions or feedback, not authoritative decisions.
- UCAS feedback is not auto-applied as the final statement — the student chooses what to keep.
- Coaching narratives are staff-facing drafts for review.
- UniGuide stays within an educational exploration flow with structured phases and guardrails.
- Provider content moderation / safety filtering applies on API calls.
9. Intellectual Property
- We have not used school IP or student content to train AI models. We do not train models.
- OpenAI and Anthropic are responsible for the training data and licensing of their foundation models.
- AI-generated outputs in VESPA should be treated as draft guidance to be reviewed and adapted by the user.
10. School liability and DPA coverage
AI use in the Platform sits within our standard data protection arrangements:
- Prompt data sent to OpenAI (and Anthropic where used) is processed under their DPAs as sub-processors of 4Sight Education Ltd.
- Our Data Processing Agreement (Schedule 2) lists these providers.
- Optional AI features form part of the educational coaching / learning service. Schools should ensure their own privacy notices to students/parents reflect AI use where required.
11. Cost
Portal AI features are included in the VESPA subscription at no additional per-query charge to the school. We manage usage through product design and provider limits.
12. Performance and Reliability
- Prompts are developed and tested by VESPA’s education / product team for UK educational context.
- Outputs remain advisory and should be reviewed by a human before consequential use.
- If an AI service is unavailable, the rest of the Platform continues to function; users see an error and can retry.
13. Transparency
- This policy is publicly available and linked from our Privacy Policy and policies index.
- We welcome questions from school Data Protection Officers and can demonstrate features on request.
14. Changes to This Policy
We may update this policy to reflect product changes, regulation, or best practice. Material changes will be communicated to partner schools. The “last updated” date indicates the most recent revision.
15. Contact
Related Documents