AI Usage Policy

Last updated: June 2026

1. Overview

VESPA Academy (“the Platform”), operated by 4Sight Education Ltd (“we”, “us”), incorporates artificial intelligence features to enhance the student coaching experience. This policy explains how AI is used within the Platform, what data is involved, and the safeguards in place.

This policy should be read alongside our Privacy Policy and Terms and Conditions.

2. AI Features in VESPA

The Platform includes three AI-powered features, all of which are optional and user-initiated:

3. How AI Is Used

• Advisory only. All AI outputs are suggestions, summaries, or feedback. No AI feature makes autonomous decisions about students, triggers automatic actions, changes grades, creates referrals, or affects access to services.
• Human-initiated. AI features only run when a user actively clicks a button (e.g. “Get Feedback” or “Generate Summary”). Nothing runs in the background, on login, or automatically.
• Human-reviewed. Coaching narratives are reviewed by tutors before being shared with students. Statement feedback is presented as suggestions for the student to consider. UniGuide recommendations are starting points for exploration, not directives.

4. AI providers and sub-processors

Optional AI features use commercial foundation models accessed via API. We do not train, fine-tune, or host AI models ourselves.

• API processing, not “sharing” for other purposes. When a user clicks to run an AI feature, the minimum data needed for that single request is transmitted over encrypted HTTPS (TLS 1.2+) to OpenAI and/or Anthropic so they can process the prompt and return a response. We do not sell, license, or disclose school or student data to third parties for advertising, profiling, or unrelated commercial use.
• Sub-processor relationship. OpenAI and Anthropic are listed as sub-processors in our Data Processing Agreement (Schedule 2) and Data Protection Policy.
• International transfers. AI API processing may occur in 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

• We do not train any AI models. We use pre-trained foundation models via API only.
• No training on school data (OpenAI). Under OpenAI’s API terms (distinct from consumer ChatGPT), data submitted via the API is not used to train or improve their models. This is set out in their API Data Usage Policy and Data Processing Agreement.
• No training on school data (Anthropic). Under Anthropic’s API commercial terms, data submitted via the API is not used to train their models by default.
• Limited retention by providers. API inputs may be retained for a short period for abuse monitoring and service operation (e.g. up to 30 days for OpenAI), then deleted. They are not retained for model training.
• No reuse by us for unrelated purposes. We do not use AI prompts or outputs to build datasets for sale, marketing, or training our own models.

6. Data minimisation

We follow the principle of data minimisation in all AI interactions:

• Only the minimum data necessary for the specific request is sent to the AI provider’s API.
• We do not send demographic data (ethnicity, gender, SEN status, disability) to AI APIs.
• We do not send student names or email addresses to AI APIs — context is provided through scores, text content, and anonymised identifiers where possible.
• Each request is independent — the model has no memory of previous requests or other students.

7. Bias and Fairness

• System prompts are written to be inclusive, encouraging, and constructive regardless of student background.
• The coaching narrative prompt explicitly requires warm, professional, specific, and actionable language.
• UCAS feedback is structured around the statement content, not the student's personal characteristics.
• OpenAI and Anthropic include safety layers designed to reduce biased or harmful outputs.
• We regularly review AI outputs against a range of student scenarios to identify and address any patterns of concern.

8. Safety and Content Controls

• All AI-generated content is clearly presented as suggestions or feedback, not authoritative decisions.
• The UCAS feedback feature requires the user to explicitly click “Add as a comment” to save generated feedback — it is not automatically applied.
• Coaching narratives are only visible to staff, who review the content before sharing with students.
• The UniGuide advisor operates within a structured multi-phase conversation flow with guardrails to keep discussions on-topic and educationally appropriate.
• Provider content moderation and safety filtering is active on API calls.
• Temperature settings are kept conservative (0.5) to reduce output variability and unpredictability.

9. Intellectual Property

• We have not used any intellectual property to train AI models. We do not train models.
• OpenAI and Anthropic are responsible for the training data and licensing of their respective foundation models.
• AI-generated outputs within VESPA (coaching summaries, feedback text) are not claimed as original works and should be treated as draft guidance to be reviewed and adapted by the user.

10. School liability and DPA coverage

There are no additional liabilities for the school regarding AI usage in VESPA beyond our standard data protection arrangements. Specifically:

• Prompt data sent to OpenAI and Anthropic is processed under their Data Processing Agreements as sub-processors of 4Sight Education Ltd.
• Data is transmitted via encrypted HTTPS, processed to generate the response, and not used by providers to train models under API business terms.
• Our Data Processing Agreement (Schedule 2) lists OpenAI and Anthropic and describes security measures.
• Schools do not need separate consent for optional AI features beyond their existing VESPA agreement, as AI processing falls within the educational coaching service — but features remain user-initiated and optional.

11. Cost

AI features are included in the VESPA subscription at no additional cost to the school. The school is not charged per AI query. Typical cost per request to 4Sight Education Ltd is under 1p, managed through token limits and on-demand-only processing.

12. Performance and Reliability

• System prompts are developed and tested against real (anonymised) scenarios by VESPA's education team to ensure outputs are appropriate, accurate, and aligned with UK educational standards.
• Outputs are advisory and always reviewed by a human (tutor or student) before any action is taken.
• We monitor API reliability and response quality, and update prompts when improvements are identified.
• AI features degrade gracefully — if the AI service is unavailable, the rest of the Platform continues to function normally. Users see a clear error message and can retry later.

13. Transparency

• This policy is publicly available and linked from our Privacy Policy.
• Schools are informed of AI features during onboarding and can discuss any concerns with their account contact.
• We are happy to provide additional information, demonstrate AI features, or discuss specific concerns with school Data Protection Officers on request.

14. Changes to This Policy

We may update this policy from time to time to reflect changes in our AI usage, the regulatory environment, or best practice. Material changes will be communicated to partner schools. The “last updated” date at the top of this document indicates the most recent revision.

15. Contact

Related Documents

• Policies & Compliance index
• Data Protection Policy
• Data Processing Agreement
• Privacy Policy
• Terms and Conditions
• Student Privacy Policy
• IT Setup Guide