- Permission-Aware
- Source-Aware
- Human-Reviewed
- Education-Specific
Draft
Prepare a starting point rather than a final decision.
Summarise
Bring together permitted information into a clearer working overview.
Explain
Make complex information easier to understand.
Structure
Turn notes or evidence into a more useful format.
Create
Help prepare educational resources and working materials.
Identify Missing Context
Show where available evidence may not support a proposed conclusion.
- Role
- School or organisation
- Learner relationship
- Case involvement
- Purpose
- Product entitlement
- Information sensitivity
- Explicit restrictions
- A teacher should not gain access to restricted safeguarding information simply by asking AI about a learner.
- A professional should not gain wider family information through an AI summary than they could access through the ordinary record.
- And access to one learner should not become a route into another learner's information.
- AI does not create new permissions.
Which Records Were Used
What information contributed to the output?
Which Version
Was the draft based on current or historic information?
When the Information Was Recorded
How recent is the context?
What Is Missing
Are important evidence areas absent?
What Was AI Generated
Which wording came from AI rather than a human source?
- Is the information accurate?
- Are the sources appropriate?
- Has important context been missed?
- Has uncertainty been represented properly?
- Does the conclusion belong to this workflow at all?
- Does another professional need to make the decision?
The purpose of AI is to reduce administrative effort around professional work – not to disguise automated judgement as human judgement.
Diagnosis
It does not diagnose autism, ADHD, dyslexia, mental-health conditions or other clinical or developmental conditions.
SEND Status
It does not automatically determine whether a learner has SEND.
EHC Plans
It does not issue an EHC plan or determine statutory entitlement.
Safeguarding
It does not make a safeguarding finding.
EOTAS
It does not determine EOTAS eligibility or suitability.
Home Education
It does not determine whether a family’s education is legally suitable.
Attendance or Statutory Decisions
It does not make attendance-order, CNIS suitability or similar statutory decisions.
Employment Decisions
It should not autonomously make recruitment, disciplinary, performance or employment decisions.
Compliance Conclusions
It does not declare an organisation legally compliant.
AI may assist authorised people with appropriate preparation. The consequential conclusion remains human.
- Summarise observations
- Organise evidence
- Prepare review notes
- Draft support-plan wording
- Explain educational information
- Identify missing evidence
- Prepare questions for further review
But an educational screening result remains educational screening.
- A family observation remains a family observation.
- A teacher observation remains teacher evidence.
- And an AI-generated summary remains a draft.
- Decide whether abuse occurred
- Determine safeguarding risk
- Decide whether to refer
- Close a safeguarding concern
- Replace DSL judgement
- Generate a statutory safeguarding conclusion
Features
Which AI tools are enabled?
Users
Which roles may use them?
Provider
Which supported AI provider is configured?
Usage
How much AI usage is permitted?
Credits or Limits
Where usage-based services apply, what limits are available?
BYOK
Where supported, schools may use an appropriate bring-your-own-key provider configuration.
Specific Workflows
A school may permit one AI feature without enabling every AI capability across the platform.
AI governance should allow schools to introduce capability deliberately rather than treating generative AI as an unavoidable platform default.
- The provider changes
- The model changes
- A model version changes
- The system prompt changes
- Product instructions change
- Safety controls change
- Available source context changes
EdiWay’s governance direction therefore includes provider, model, prompt and policy versioning for governed AI workflows. Where advanced controls are supported, changes can be subject to testing and approval rather than assuming a new model is equivalent to the previous one.
- Missing records
- Limited evidence
- Old information
- Conflicting sources
- Incomplete context
- Unavailable information
The system should not invent artificial certainty simply because a user asks for a definitive answer. Where confidence cannot be meaningfully calibrated, the platform should explain the limitation rather than present a percentage that gives false reassurance.
Age
Is the interaction appropriate for the learner?
SEND
Does the design work appropriately for different communication and learning needs?
Filtering
Can harmful or inappropriate content be reliably restricted?
Context
Is the AI staying within the educational task?
Monitoring
Can relevant safety events be identified through an appropriately governed process?
Human Support
Does the experience direct the learner towards real people when human support is needed?
Development
Is AI helping the learner think rather than simply doing all of the thinking for them?
Learner-facing AI requires a child-centred safety approach – not simply the same assistant shown through a smaller screen.
- Start with a prompt or hint.
- Ask what the learner already understands.
- Encourage an attempt.
- Explain the next step.
- Provide more support as needed.
Rather than automatically replacing the learner’s own thinking with a complete solution. The objective should be:
- Artificial flattery designed to influence decisions
- False certainty
- Fear-based prompts
- Guilt
- Social-pressure messages
- Hidden commercial steering
- Engagement tricks designed simply to increase AI use
- Dark patterns
- AI-generated pressure to purchase additional services
AI should help a user accomplish a legitimate task. It should not manipulate them into staying, spending or agreeing.
- Why information is being processed
- Which information is necessary
- Who is authorised
- Which provider processes it
- Where processing occurs
- Relevant contractual safeguards
- Retention
- Logging
- User transparency
- Appropriate risk assessment
- Sensitive information
The exact obligations depend on the organisation, workflow, provider and data involved. EdiWay should therefore not claim that activating AI automatically makes a school’s AI use compliant.
- User
- School or organisation
- Learner context
- AI tool
- Provider/model
- Source context
- Generation
- Review state
- Relevant revisions
- Usage
- Material approvals
This can help schools understand how AI is being used and investigate problems when they arise. Logging itself must also remain proportionate and appropriately protected.
Permission Failure
Can the AI retrieve information from another learner, role or school?
Prompt Manipulation
Can a user persuade the system to ignore restrictions?
Harmful Content
Can learner-facing AI produce or retrieve inappropriate material?
Missing Evidence
Does the AI invent information when source context is incomplete?
Model Changes
Does a newer model still behave within the intended boundary?
Sensitive Information
Can protected information escape through summaries or generated documents?
Human Review
Can a consequential workflow bypass the intended review stage?
Provider Failure
What happens when the external AI provider is unavailable or returns an invalid response?
AI safety is not demonstrated because normal prompts work correctly. The failure paths matter too.
- Stated purpose
- Educational use cases
- Filtering
- Monitoring and reporting
- Security
- Privacy and data protection
- Intellectual property
- Design and testing
- Governance
- Cognitive development
- Emotional and social development
- Mental health
- Manipulation
EdiWay’s governance direction is intentionally consistent with principles such as bounded purpose, permission control, evidence-based claims, safe design, human accountability and proportionate learner safeguards.
This does not mean EdiWay claims DfE certification, endorsement or approval.
The DfE standards apply to England, while EdiWay serves a wider UK education context where jurisdiction and organisational policy must also be considered.
1. Define the Task
What is AI being asked to help with?
2. Establish the User
Who is making the request?
3. Check Permission
What information can that user access for this purpose?
4. Limit the Context
Provide only the information needed for the task.
5. Generate
Create a draft or assistance output.
6. Preserve Relevant Provenance
Keep appropriate information about the evidence and AI process.
7. Surface Limitations
Identify missing or incomplete context.
8. Human Review
An authorised user checks the result.
9. Approve, Edit or Reject
AI does not approve itself.
10. Retain Appropriate Governance Evidence
Record material use according to the workflow.
- Supported capability
- Integrated foundation
- Controlled or policy-dependent capability
- Future capability
Governance should apply to the claims made about AI as well as the AI itself.
No.
EdiWay AI is designed to assist with drafting, summarising, explaining and organising information.
Consequential professional and statutory decisions remain human decisions.
No.
AI should remain inside the same role, organisation, learner, relationship, purpose and sensitivity boundaries that apply to the user.
No.
AI does not create new permissions.
No.
It may help organise permitted educational evidence, but it does not diagnose autism, ADHD, dyslexia or other conditions.
AI may assist with permitted working drafts where configured, but it does not issue an EHC plan or make the statutory decision.
No.
Safeguarding decisions remain with appropriately authorised people.
Broad AI access to safeguarding information is excluded by default.
They should not be.
Consequential AI output remains subject to human review and the relevant workflow.
EdiWay’s AI direction includes school-level controls over relevant features, provider and usage configuration where supported.
Bring-your-own-key provider configuration forms part of the supported AI direction where enabled and appropriately configured.
EdiWay’s governed AI architecture includes source and provenance controls, although the exact rendered source experience may vary by AI workflow as broader assurance continues.
It should not.
Governed AI should identify limited or missing evidence rather than manufacture certainty.
Learner-facing AI should only be enabled through appropriate age, safety, permission and product controls.
The safer educational direction is to encourage thinking, attempts, hints and progressive support rather than replacing the learner’s work by default.
No such endorsement should be implied.
EdiWay uses relevant DfE generative AI safety principles as an important design reference for education AI.
No.
Schools remain responsible for their own governance, policies, configuration and lawful use.
Trust and Governance
Understand EdiWay’s wider governance model.
Permissions, Consent and Information Sharing
See how access is limited by relationship, purpose and information scope.
Data Security and Governance
Explore security, audit and access controls.
Safeguarding Boundaries
Understand the stronger boundary around safeguarding information.
Privacy and Data Protection
Understand how personal information and AI fit within the wider privacy model.
Data Retention and Subject Rights
Understand retention, account lifecycle and information rights.
Implementation, Migration and Product Assurance
See how EdiWay distinguishes product capability from deployment and assurance.
- Use AI to draft.
- Use it to summarise.
- Use it to organise evidence.
- Use it to prepare resources.
- Use it to make repetitive information work easier.
- But keep permissions around the information.
- Keep sources behind the output.
- Keep sensitive records protected.
- Keep limitations visible.
- Keep children at the centre of learner-facing safety.
- And keep consequential decisions with the people responsible for making them.